20190704_AGIT_Georaster_ImageryData_KPatenge

Copyright © 2019, Oracle and/or its affiliates. All rights reserved. |
Imagery Data with
Oracle Spatial Technologies
Karin Patenge | karin.patenge@oracle.com | @kpatenge
Principal Solution Engineer | BU Cloud & Core Technologies
Oracle Deutschland B.V. & Co. KG
AGIT 2019 | July 3rd - 5th, 2019 | Salzburg, AT
oracle-spatial.blogspot.com | blogs.oracle.com/oraclespatial | blogs.oracle.com/bigdataspatialgraph

Today´s focus on Database-centric
Raster Data Processing & Analysis
2

Note
• For processing raster data using Hadoop and Spark please refer to workshop (HOL)
materials available for
– Applying Hadoop Spatial Analysis To Big Data
download.oracle.com/otndocs/products/spatial/pdf/BD-Spatial-HOL-Hadoop.zip
– Applying Spark Spatial Analysis To Big Data
download.oracle.com/otndocs/products/spatial/pdf/BD-Spatial-HOL-Spark.zip
• Further Links
– Big Data Spatial and Graph Product and Technology Pages
www.oracle.com/technetwork/database/database-technologies/bigdata-spatialandgraph/overview/index.html
www.oracle.com/database/big-data-spatial-and-graph/
– Whitepaper for Spatial Features and Big Data Spatial and Graph
download.oracle.com/otndocs/products/bigdata-spatialandgraph/bdsg-spatial-feature-overview.pdf
– Big Data Spatial and Graph Blog
blogs.oracle.com/bigdataspatialgraph/
3

Where to get slides and scripts from?
• Slides can be downloaded from
www.slideshare.net/kpatenge
• Scripts & much more can be downloaded from
tinyurl.com/oraclespatialws2019
4

5
Sentinel 2 Data
& others
scihub.copernicus.eu/dhus/#/home earthexplorer.usgs.gov/

Copyright © 2019, Oracle and/or its affiliates. All rights reserved. | 6
Access via USGS EarthExplorer

Enterprise Database-centric Approach
Move algorithms
closer to data
Parallel
Processing
Raster Algebra
Language
16
Data Management & Data Processing & Data Analysis inside the Oracle Database
In-
Database
Analytics
Engine

Oracle Database as Multimodel Database
• Oracle Database supports multiple models
– Relational, In-memory, Sharded
– Document Store
• JSON
• XML
• Text
• OLAP
– Spatial Database
– Graph Database and Triple Store
• Oracle Database support multiple languages
and access protocols

Oracle Spatial and Graph
Three major features
Spatial Property Graph RDF Graph

Oracle Spatial and Graph
Location and graph analysis with secure storage for enterprise data
Topologies
3D / LiDAR
Raster
Networks
Web Services (OGC)Geocoding RoutingDeployable Services
RDF Graphs Property Graphs
Mapping
Polygons
Lines
Points
Location Tracking
(Geofencing)

General Process Flow
Raster Data
Load, Storage
& Manage-
ment
Raster Data
Querying &
Manipulation
Raster
Algebra &
Analytics
20

Demo Environment
• Oracle Database 19c deployed on Oracle Cloud Infrastructure (OCI)
– cloud.oracle.com
– Trial Cloud Accounts (1 month + $300 credits) available via  or via an Oracle Solution Engineer
(2 months + $500 credits, no credit card)
– Integration with GDAL via PL/SQL Package SDO_GEOR_GDAL and via $ORACLE_HOME/md/gdal
• Client Software
– Oracle SQL Developer (free)
– SSH Client (PuTTY, MobaXterm, or similar)
• Most useful skills
– SQL, PL/SQL, Shell scripting
21

22
GeoRaster
11

GeoRaster Slides based on Oracle Spatial
Workshop created and updated by
Albert Godfrind (Oracle Corporation)
23

Topics
24
Concepts
Storage Model
Loading and exporting Rasters
Raster Viewer and ETL
Raster Algebra and Analytics
Mosaicing and other Image Processing
What´s New with Oracle DB 19c
1
2
3
4
5
6
7

Georaster Concepts

Raster/Vector Data Differences
Vector Data Raster data
• Precise coordinates define a point,
line, or polygon
• Sparse representation of data
(requires less storage)
• Good for representing map features
• Good for describing connectivity
(topology)
• One or more cells approximate a point, line,
or polygon
• Requires storage of a complete matrix of
cells
• Good for representing and analyzing large,
filled structures
• Easily drawn, as each cell can correspond to
one or more pixels
• Typically requires much more storage space

Raster/Vector Data Differences
Raster Data
-74.1651749, 41.339141
-74.1651749, 41.339141,
-73.4284481, 40.678193,
-72.9792214, 41.686228
-74.1651749, 41.339141,
-74.1651749, 39.559004,
-72.9792214, 39.559004,
-72. 9792214, 41.339141
Vector CoordinatesVector Data

Raster Data and Cell Size
Coarser resolution
Finer resolution

What is a Raster ?
• Two dimensional array of regularly spaced elements (pixels or
cells)
– Orthophotos
– Remote Sensing
– Gridded data (raster GIS)
• Digital images can be composed of one or more bands
– Bands often represent an interval of wavelengths along the electromagnetic
spectrum
– Band data can be simultaneously recorded
• Image data is collected by a variety of technologies
– Satellite remote sensing
– Airborne photogrammetry
– Sonar
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679

Cells and Bands
• Cell data is the value associated with the cells/pixels
– Each cell has a value in a raster
– Each value is associated with a cell/pixel
– In an RGB image, each cell has 3 values: one each to represent
the intensity of red, green, and blue
• In GeoRaster, all cells associated with a 2-D set of rows
and columns in a plane is a band
– An RGB image will have 3 bands
– A multispectral image with 4 channels will have 4 bands in
GeoRaster
– Grid data with 4 bands will have 4 bands in GeoRaster
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679

Multi band Images
TM Band 2 TM Band 3 TM Band 4
TM Bands 4+3+2
Some bands
accentuate
different features

Resolution Pyramid
Typically increases
storage cost by 33% to
40%
Gradually simplified
copies of the raster
Step down the pyramid
when zooming in
Pyramid
Level 0
Pyramid
Level 1
Pyramid
Level 2

Blocking
• A GeoRaster image can be composed of an extremely large number of cells
• It is more efficient in terms of storage and retrieval to break large images into smaller blocks
• In GeoRaster, users/applications can determine how data is blocked
– Specify rows, columns, and optionally bands
n x m blocks

Cell Depth
• Specifies the size of the cells or pixels
• Integers can be 8bits, 16bits or 32bits,
signed or unsigned
• Default for imagery is 8BIT_U
1BIT
2BIT
4BIT
8BIT_U
8BIT_S
16BIT_U
16BIT_S
32BIT_U
32BIT_S
32BIT_REAL
64BIT_REAL

JPEG Compression
• Lossy compression
• Use for imagery
• For rasters with cellDepth=8BIT_U and
no more than 4 bands per block
• Control the compression level using the
quality parameter
– 0 (max compression) to 100 (no compression)
– Default quality is 75
JPEG-2000 Compression
• Lossy compression but compresses
much better than JPEG
• Use for imagery
• For rasters with cellDepth=8BIT_U or
16BIT_U
• Control the compression level using
the ratio parameter.
– A ratio of 20 means a reduction of 20:1, i.e.
5% of the input.
Compression for Imagery

Compression for Grids
DEFLATE Compression
• Lossless compression
• Use for grids, terrain models, etc
• Uses the ZLIB format

The Impact of Compression
• Consider one TIFF file of 53 MB (uncompressed)
• When compressed as JPEG-F it is reduced to 4.4 MB
• This a compression ratio of 1:12
• Add 50% to accommodate the pyramid (approx 2.1 MB)
• Total storage will be approx 6.5 MB
• This is now a compression ratio of 1:8

With Database Compression
Without Database Compression
The Impact of Compression
Files Database (compressed)
Compression
Average
file size
(MB)
Number
of files
Total
size (MB)
Total
size
(GB)
Base
size
(MB)
Pyramid
Size
(MB)
Total
Size
(MB)
Ratio
with Files
Uncompressed 52,9 10 000 529 000 517 44 083 23 364 67 448 0,13
Compressed 4,4 10 000 44 083 43 44 083 23 364 67 448 1,53
Files Database (uncompressed)
Compression
Average
file size
(MB)
Number
of files
Total
size (MB)
Total
size
(GB)
Base
size
(MB)
Pyramid
Size
(MB)
Total
Size
(MB)
Ratio
with Files
Uncompressed 52,9 10 000 529 000 517 529 000 211 600 740 600 1,4
Compressed 4,4 10 000 44 083 43 529 000 211 600 740 600 16,8

Value Attribute Table
• A Value Attribute Table (VAT) is
used to map the stored numeric
values to the meaning of that
value
• An example value attribute table
for geological raster data
44
Cell Value Geological Layer
1 Quaternary
2 Tertiary
3 Paleocene-Cretaceous
4 Mesozoic
5 Gondwana
6 Early Palaeozoic
7 Proterozoic
8 Early Proterozoic
9 Archaean
0 Blank Cell (no data)
Stored cell values
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679

Colormap Table
• A COLORMAP table is used to
map the stored numeric values
to the display characteristics of
that value
• An example COLORMAP table
for geological raster data
45
Cell Value Red Green Blue
1 255 255 0
2 82 123 67
3 142 230 98
4 96 121 228
5 145 231 243
6 255 51 0
7 203 188 224
8 195 135 75
9 204 102 255
0 0 0 0
6 1 1 1 1 661
1 3 8 7 9 197
3 1 8 3 3 195
3 3 3 9 8 197
0 3 3 3 9 019
0 8 8 9 9 001
0 0 2 9 1 000
2 5 4 9 1 679
Stored cell values

Georeferencing
• Georeferencing establishes the relationship between cell coordinates of
GeoRaster data and real-world ground coordinates or other local
coordinates
• Georectification is the process of assigning ground control points (GCPs) to
digital images and processing the image to better map it to the surface of
the Earth
• GeoRaster supports affine transformations that georeference two
dimensional raster data
– The implementation in GeoRaster is known as the GeoRaster Spatial Reference
System

Georeferencing
Restaurant
National
Park
X
Y
GeoRaster Spatial
Reference System (SRS)
row
column
Raster Data
Cell Coordinate System
Geographic Entities
Model (ground) space
Model (ground) coordinate system

Storage Model

Enabling GeoRaster
• Georaster is not enabled by default !
• Remember to enable it:
• For multi-tenant environments: do this in each PDB!
SQL> DESCRIBE SDO_GEORASTER
ERROR:
ORA-04043: object sdo_georaster does not exist
SQL> EXECUTE MDSYS.enableGeoRaster;

Storage Model
• Separates logical from physical structures
• Logical structures
– Tables containing an SDO_GEORASTER column
– Contains raster metadata and footprint
– Also contains a pointers to an RDT
• Physical structures
– “Raster data tables” or RDTs
– Contain raster blocks
– Can be very large
– Structure defined in SDO_RASTER type

Logical structures
Contains raster metadata and
footprint
Also contains a pointers to one or
more RDT’s
Physical structures
“Raster data tables” or RDTs
Contain raster pixels
Can be very large
Storage Model: Separates Logical from Physical structures
raster 11
raster 12
raster 13
raster 14
raster 11 blocks
raster 12 blocks
raster 13 blocks
raster 14 blocks
raster 15 blocks
raster 16 blocks
raster 15
raster 16

Storage Model
raster 11
raster 12
raster 13
raster 14
raster table raster 11 blocks
raster 12 blocks
raster data table 1
raster 13 blocks
raster 14 blocks
raster 15 blocks
raster 16 blocks
raster 15
raster 16
raster data table 2

Storage Model
raster 11
raster 12
raster 13
raster 14
raster table 1
raster 11 blocks
raster 12 blocks
raster data table
raster 13 blocks
raster 14 blocks
raster 15 blocks
raster 16 blocks
raster 15
raster 16
raster table 2

A more complex example: not recommended!
Storage Model
raster 11
raster 12
raster 13
raster 14
raster table 1
raster 21
raster 22
raster 23
raster 24
raster table 2
raster 11 blocks
raster 21 blocks
raster 22 blocks
raster 13 blocks
raster 24 blocks
raster 12 blocks
raster 14 blocks
raster 23 blocks
raster data table 1
raster data table 2
raster data table 3

A Look Inside Raster Blocks
• Binary integers (16 bit, 32 bit) are stored in big-endian encoding
• Floating points (32 bit and 64 bit) are in IEEE 754 encoding
• Small binary integers (1 bit, 2 bit, 4 bit) fill bytes: a byte will therefore
contain 8x1-bit cells, or 4x2-bit cells, or 2x4-bit cells.
– Bytes are filled from “left” to “right”, i.e. from high-order bits to low-order bits

Importing and Exporting Rasters

Loading Rasters: Flow of Operation
Create the raster table(s)
Load from the database (the
"pull" method)
Create the raster data table(s)
Initialize the raster
Load into the database (the
"push" method)
Georeferencing
Set Georeferencing Parameters
Set Model SRID
Generate Spatial Extent
Generate resolution pyramid
Create spatial index

CREATE TABLE US_RASTERS(
GEORID NUMBER PRIMARY KEY,
SOURCE_FILE VARCHAR2(80),
DESCRIPTION VARCHAR2(32),
GEORASTER SDO_GEORASTER
);
Creating Raster Tables
• Use the SDO_GEORASTER type
• Combine with any other
attributes
RASTERTYPE NUMBER
SPATIALEXTENT SDO_GEOMETRY
RASTERDATATABLE VARCHAR2(32)
RASTERID NUMBER
METADATA SYS.XMLTYPE

Creating Raster Data Tables
• Use the SDO_RASTER type
• Control the storage of LOBs (use SECUREFILE lobs)
CREATE TABLE RDT_01 OF SDO_RASTER (
PRIMARY KEY (
RASTERID, PYRAMIDLEVEL, BANDBLOCKNUMBER,
ROWBLOCKNUMBER, COLUMNBLOCKNUMBER
)
)
LOB(RASTERBLOCK) STORE AS SECUREFILE (
STORAGE (INITIAL 1024G)
NOCACHE NOLOGGING
);
RASTERID NUMBER
PYRAMIDLEVEL NUMBER
BANDBLOCKNUMBER NUMBER
ROWBLOCKNUMBER NUMBER
COLUMNBLOCKNUMBER NUMBER
BLOCKMBR SDO_GEOMETRY
RASTERBLOCK BLOB
RDTs must have unique names (across schemas)
RDTs not listed in USER_TABLES but in USER_OBJECT_TABLES

After all loading is completed …
• Re-enable caching and logging on the RDTs
ALTER TABLE RDT_01
MODIFY LOB(RASTERBLOCK)
(
CACHE
LOGGING
);

Initializing Rasters
• Specify the storage location for the blocks
– Name of raster data table
• Will be created if it does not exist!
– Unique id in that table
• If no id provided, one will be assigned automatically
INSERT INTO US_RASTERS (GEORID, GEORASTER) VALUES (1, SDO_GEOR.INIT('RDT_01',1));
This is generally done by the loading tool
Raster
Data Table Raster id

• Oracle tools
– Java command line
– Raster viewer/importer/exporter
– MapBuilder
– Java API
• Open Source
– GDAL
• Commercial tools
– FME
• Database-Driven
– Use PL/SQL
– SDO_GEOR.ImportFrom()
– SDO_GEOR_GDAL.TRANSLATE()
• Georaster ETL
– Choose files to load
– Set parameters
– Uses GDAL for the actual loading
– Parallel loading
Importing Rasters

Importing using the Java command line
java -Xms900M -Xmx900M GeoRasterLoader
127.0.0.1 spatialdb 1521
scott tiger
thin 32 T
us_rasters georaster
"blocking=true,blocksize=(512,512,3),
spatialExtent=true,compression=jpeg-f"
"sf1.tif,1,RDT_01,sf1.tfw,26943"
JDBC connection
parameters
Table and column
Load parameters
Source file
and TFW
Raster id and
data table
Projection

Database-driven Import: importFrom()
declare
g sdo_georaster;
begin
select georaster into g from us_rasters
where georid = 1 for update;
sdo_geor.importFrom(
g,
'blocksize=(512,512)',
'TIFF', 'file', '/usr/rasters/sf1.tif',
'WORLDFILE', 'file', '/usr/rasters/sf1.tfw',
);
update us_rasters
set georaster = g
where georid = 1;
end;
Read Raster
For update
Import from file
Specify blocking and
compression
Update Raster

Database-driven Import
Grant file access
Prior to load: must grant access to the files to load to the loading user
• Grant access to individual files:
• Grant access to all files in a directory :
• Grant access to all files on the server :
SQL> call dbms_java.grant_permission('SCOTT',
'java.io.FilePermission', '/usr/data/rasters/sf1.tif','read');
'java.io.FilePermission', '/usr/data/rasters/sf1.tfw','read');
'java.io.FilePermission', '/usr/data/rasters/*','read');
'java.io.FilePermission', ‘<<ALL FILES>>','read');
SQL> call dbms_java.grant_permission('MDSYS','java.io.FilePermission', '<<ALL FILES>>','read');

Exporting using the Java command line
java -Xms900M -Xmx900M GeoRasterExporter
127.0.0.1 orcl101 1521
georaster georaster
thin 32 T
us_rasters georaster
"1,UK_RASTERS_RDT_1"
"JPEG"
raster_1_p0_0_0.500_500.jpg
"pLevel=0,cropArea=(0,0,500,500)"
JDBC connection
parameters
Table and column
Raster ID
Output format
Output file name
Reformatting

Database-driven Export: exportTo()
declare
g sdo_georaster;
begin
select georaster into g from us_rasters
where georid = 1;
sdo_geor.exportTo(
g,
'pLevel=0,cropArea=(0,0,500,500)',
'TIFF', 'file', '/usr/rasters/sf1s.tif',
'WORLDFILE', 'file', '/usr/rasters/sf1s.tfw',
);
end;
Read Raster
Export to file
Specify reformatting
parameters

GeoReferencing
• Set the georeferencing parameters
– Usually set automatically when loading: extracted from the GeoTIFF tags or from the
TFW file a
• Set the coordinate system of the raster
– Usually set automatically when loading: extracted from the GeoTIFF tags
– Or specified as parameter when loading
• Compute the spatial extent of the raster
– Usually computed automatically when loading

GeoReferencing
Set SRID and Spatial Extent
declare
geor sdo_georaster;
begin
select georaster into geor from us_rasters
sdo_geor.setModelSRID(geor, 40978);
geor.spatialExtent :=
sdo_geor.generateSpatialExtent(geor);
update us_rasters
set georaster = geor
where georid = 1;
end;
Read Raster
for update
Set the SRID
Set the spatial
footprint
Update Raster

Generating a resolution pyramid
declare
geor sdo_georaster;
begin
select georaster into geor from us_rasters
sdo_geor.generatePyramid(
geor,
'resampling=nn rlevel=4'
);
update us_rasters
set georaster = geor
where georid = 1;
end;
Generate 4-level
pyramid
Read Raster
for update
Update Raster

Pyramiding parameters
• Resampling Modes
– NN: Value of the nearest neighbor cell (used by default)
– BILINEAR: Distance-weighted average of the 4 nearest cells
– AVERAGE4: Simple average of the 4 nearest cells
– AVERAGE16: Simple average of the 16 nearest cells
– CUBIC: Cubic convolution of the 16 nearest cells
• Number of levels
– If not specified, pyramid levels are generated until the smaller of the number of rows
or columns is between 64 and 128.

Using GDAL

Using GDAL
• Open source library and tools
• Supports many formats
• Sources available from http://www.gdal.org/
• Version 2.1.3 included in $ORACLE_HOME/md/gdal (of Oracle DB 18c)
• Also available online inside file GeoRasterETL.zip from
– http://www.oracle.com/technetwork/indexes/samplecode/spatial-1433316.html
– gdal_linux.x64.zip
– gdal_windows.x64.zip
• Use the proper version for your platform

Using GDAL
• You can also download Windows pre-built binaries from
http://trac.osgeo.org/osgeo4w/
– Packaging of a number of open source tools for Windows (GDAL/OGR, GRASS,
MapServer, OpenEV, uDig, QGIS, …)
– Download an installer and select the packages to install
• Other source: http://www.gisinternals.com/sdk/
– Daily builds of stable and development versions for multiple windows variants
(includes WIN64).
– Includes windows installers

Using GDAL
• To setup
– Unzip the archive anywhere
– Add /bin to your path and /lib to your library path
– Set GDAL_DATA to /data (optional)
• Main commands for raster data
– gdalinfo = get information about a file (or spatial table)
– gdal_translate = copy from one format to another (import/export)
• Documentation
– http://www.gdal.org/
– http://www.gdal.org/ogr_formats.html
– http://www.gdal.org/drv_oci.html

Pre-packaged GDAL
• Linux and Windows only
• Installed in $ORACLE_HOME/md/gdal
• Need to setup your environment to use it
Confidential – Oracle
89
export GDAL_HOME ${ORACLE_HOME}/md/gdal
export GDAL_DATA ${GDAL_HOME}/data
export GDAL_DRIVER_PATH ${GDAL_HOME}/lib/gdalplugins
export PATH ${GDAL_HOME}/bin:${PATH}
export LD_LIBRARY_PATH ${GDAL_HOME}/lib:${LD_LIBRARY_PATH}

GDAL supported Formats
• See complete list here:
http://www.gdal.org/formats_list.html
• Some formats are “built-in”, some use
plugins
• Find details about one format:
• Useful to find out the syntax of
parameters specific to this format
AAIGrid (rw): Arc/Info ASCII Grid
AIG (ro): Arc/Info Binary Grid
...
EIR (ro): Erdas Imagine Raw
ELAS (rw+): ELAS
ERS (rw+): ERMapper .ers Labelled
ESAT (ro): Envisat Image Format
...
GTiff (rw+): GeoTIFF
GXF (ro): GeoSoft Grid Exchange Format
...
GeoRaster (rw+): Oracle Spatial GeoRaster
...
HFA (rw+): Erdas Imagine Images (.img)
...
JPEG (rw): JPEG JFIF
...
VRT (rw+): Virtual Raster
XPM (rw): X11 PixMap Format
gdalinfo --formats
gdalinfo --format georaster
Oracle Georaster
plugin is installed

Main GDAL Utilities
• gdal_translate
– Main tool to convert a raster from one format to another (like TIFF to JPEG)
– Used to import rasters into Oracle as well as to export from Oracle
– Can be used to change the structure of a raster from BSQ to BIP
– Can also be used to extract the georeferencing information from a GeoTIFF raster into
a TFW file
– Many other conversions are possible, such a breaking a very large file into a set of
smaller files or fusing many into a single large raster (mosaicing)

Other GDAL Utilities
• gdalinfo
– Prints out information about a raster
– Useful to find out any georeferencing and projection information as well as the
structure of a raster
• gdaladdo
– Adds overviews to a raster, i.e. generate a resolution pyramid
• gdalwarp
– Reprojects a raster to a different coordinate system

Importing a raster from a file
gdal_translate -of georaster sf1.tif
georaster:scott/tiger@spatialdb,us_rasters,georaster
-co blockxsize=512 -co blockysize=512 -co blockbsize=3
-co interleave=bip -co srid=26943
-co "insert=(georid, source_file, georaster) values (1, 'sf1.tif',
sdo_geor.init('us_rasters_rdt_01'))"
-of georaster Use Oracle Georaster as output
sf1.tif Name of input TIFF file
georaster:scott/tiger@spatialdb Database connection
us_rasters,georaster Destination table and column
-co block%size Block size in x, y and b (band)
-co SRID=26943 Coordinate system of the raster
-co interleave Interleaving (default is band sequential)
-co insert The parameters of an insert statement to insert a row in the raster table
-co genpyramid=nn Generate the resolution pyramid after the load has completed

Using Optimal Block Size
• Blocks may include padding
– For example: raster is 4299x4299
– Blocksize 512x512 is not optimal – produces padding and wasted space
• This uses a block size of 430x430 – no or little padding
-co blocking=optimalpadding
-co "insert=values (1, 'sf1.tif', 'Aerial photo San Francisco 1',
sdo_geor.init('us_rasters_rdt_01', 1),null,null)"

Compress while Loading
• Much more efficient than loading, then compressing later
-co compress=jpeg-f

Mosaic while Loading
• Load a set of raster files into a single georaster object
• More efficient than loading, then mosaicing later
• First, define a “virtual data set” (VRT)
– Lists all the tif files to combine
• Then load the “virtual data set”
gdalbuildvrt sf.vrt *.tif
gdal_translate -of georaster sf.vrt
-co "insert=values (1, 'sf.vrt', 'Aerial photo San Francisco',
sdo_geor.init('us_rasters_rdt_01’),null,null)"

Reproject while Loading
• Use gdalwarp instead of gdal_translate
• -t_srs specifies the coordinate system to reproject to
• -a_srs specifies the coordinate system of the input file
– Needed only if it cannot be determined from the geotiff tags in the input file
gdalwarp -of georaster sf1.tif
-co interleave=bip
-overwrite
-t_srs EPSG:4326

Reproject, Compress and Mosaic while Loading
• Define the “virtual data set” (VRT)
– Lists all the tif files to combine
• Then process the “virtual data set”
gdalbuildvrt sf.vrt *.tif
gdalwarp -of georaster sf.vrt
-co interleave=bip
-co compress=jpeg-f
-overwrite
-t_srs EPSG:4326
-co "insert=values (1, 'sf.vrt', 'Aerial photo San Francisco',
sdo_geor.init('us_rasters_rdt_01’),null,null)"

Native JPEG 2000 Compression
• The same as JP2 file. Stored in a single BLOB. Image can be tiled
• High compression ratio. For the same quality, 2-7 times smaller than JPEG
image
• Support both lossless and lossy compressions
• No separate pyramids. Pyramid level is limited by tile dimension sizes
• Support 8bit and 16bit integer images
• No limit on number of bands
• Support very large size image. Size is limited by memory and max number
of tiles (max_mem_size / 20 * 65535)
99

Compress as JPEG2000
• No need to specify block sizes
• Compressing automatically generates the pyramid
• Specify the compression level using the JP2_QUALITY parameter
– represents the percentage in size reduction
– for example: 5 means that the raster is reduced to 5% of the input file
– this is a compression ratio of 100/5 = 20:1
• Default quality is 25 (= ratio 4:1)
georaster:scott/tiger@orcl111,us_rasters,georaster
-co compress=jp2-f

Compress as JPEG2000
• GDAL build must contain a JPEG2000-capable plugin
– JP2OpenJPEG (Open source) or JP2ECW (ERDAS SDK)
– Load fails if none is present:
• Plugin must allow compression.
• Specifying JP2_QUALITY=100 gives a lossless compression
– Still smaller than the original file
ERROR 1: Unable to copy JPEG2000 image into GeoRaster dataset.
No JPEG2000 capable driver (JP2OPENJPEG, JP2ECW, JP2MRSID, etc...) is available."
ERROR 1: Unable to copy JPEG2000 image into GeoRaster dataset.
One or several JPEG2000 capable drivers are available but the file could not be opened
successfully.

Loading JPEG2000 raster files
• The process copies the original JP2 "as-is"
– Includes the existing pyramid
– Very fast
• Do not specify any quality (the quality of the input is retained)
gdal_translate -of georaster sf1.jp2
-co compress=jp2-f
-co "insert=values (1, 'sf1.jp2', 'Aerial photo San Francisco 1',

Reading ECW rasters
• Requires the ECW GDAL plugin
– Provided with our build, but not enabled
– To enable, unzip the plugin file
• Requires the Erdas ECW SDK
– Download from Erdas’ download site
– Install and configure
$ gdalinfo --formats | grep ECW
ECW -raster- (rw+): ERDAS Compressed Wavelets (SDK 5.0)

Enable ECW GDAL Plugin
$ cd $GDAL_HOME/lib/gdalplugins
$ ls -l *ECW*
-rw-rw-r--. 1 spatial spatial 58935 Nov 12 2015 gdal_ECW_JP2ECW.zip
$ unzip gdal_ECW_JP2ECW.zip
Archive: gdal_ECW_JP2ECW.zip
inflating: gdal_ECW_JP2ECW.so
$ ls -l *ECW*
-rwxrwxr-x. 1 spatial spatial 152352 Nov 12 2015 gdal_ECW_JP2ECW.so
-rw-rw-r--. 1 spatial spatial 58935 Nov 12 2015 gdal_ECW_JP2ECW.zip

Download ERDAS ECW SDK
• http://download.hexagongeospatial.com
• Select product: “ERDAS ECW/JPEG2000 SDK”
• Select the version for your platform
– Windows, MacOS, Linux
• “Download now”
• Register, then download starts (around 200 MB)
• Or just grab directly:
$ wget http://go.hexagongeospatial.com/ECWJP2SDK5.3.0LINUX

Download
ERDAS ECW
SDK

Install ERDAS ECW SDK
• Unzip the SDK
• Make it executable
• Run the installer
$ unzip erdas-ecw-sdk-5.3.0-linux.zip
Archive: erdas-ecw-sdk-5.3.0-linux.zip
inflating: ERDAS_ECWJP2_SDK-5.3.0.bin
$ chmod +x ERDAS_ECWJP2_SDK-5.3.0.bin
$ ./ERDAS_ECWJP2_SDK-5.3.0.bin

ERDAS ECW JPEG2000 SDK 5.3.0 Install
************************************
Please select the appropriate license type to deploy ...
No License Fee Required
********************************
Enter 1 for "Desktop Read-Only Redistributable"
Enter 2 for "Mobile Read-Only (Local decoding restriction apply)”
Paid Licensees
********************************
Enter 3 for "Desktop Read-Write Redistributable"
Enter 4 for "Server Read-Only End User"
Enter 5 for "Server Read-Only Redistributable"
Enter 6 for "Server Read-Write Redistributable”
[1, 2, 3, 4, 5, or 6]
1
Displaying End User License Agreement ...
ERDAS ECW/JP2 Desktop Read-Only Redistributable SDK License Agreement
...
Do you accept this License Agreement? [yes/no]
yes
Unpacking the ERDAS ECW JPEG2000 SDK 5.3.0 package to /home/spatial/hexagon/ERDAS-ECW_JPEG_2000_SDK-5.3.0/Desktop_Read-Only
Please wait …
Done.
Result goes to: $HOME/hexagon/ERDAS-ECW_JPEG_2000_SDK-5.3.0/Desktop_Read-Only

• Result goes to :
• The shared libraries are in lib/x64/release :
• Add to library path
$HOME/hexagon/ERDAS-ECW_JPEG_2000_SDK-5.3.0/Desktop_Read-Only
$ cd $HOME/hexagon/ERDAS-ECW_JPEG_2000_SDK-5.3.0/Desktop_Read-Only
$ ls /lib/x64/release
libNCSEcw.a libNCSEcw.so libNCSEcw.so.5.3.0
libNCSEcwJNI.so.5.3.0 libNCSEcw.so.5 libNCSEcw.so.debug
libNCSEcwJNI.so.debug libNCSEcw.so.5.3
$ export LD_LIBRARY_PATH=/opt/erdas/ERDAS-ECW_JPEG_2000_SDK-5.3.0/Desktop_Read-
Only/lib/x64/release:$LD_LIBRARY_PATH

Get information about a raster file
gdalinfo sf1.tif
Driver: GTiff/GeoTIFF
Files: ........data11-georastertiffsf1.tif
........data11-georastertiffsf1.tfw
Size is 4299, 4299
Coordinate System is:
LOCAL_CS["NAD83 / California zone 3", …]
Origin = (1828467.919737499900000,646446.278891499970000)
Pixel Size = (0.304800999999999,-0.304800999999999)
Metadata:
AREA_OR_POINT=Area
Image Structure Metadata:
INTERLEAVE=PIXEL
Corner Coordinates:
Upper Left ( 1828467.920, 646446.279)
Lower Left ( 1828467.920, 645135.939)
Upper Right ( 1829778.259, 646446.279)
Lower Right ( 1829778.259, 645135.939)
Center ( 1829123.089, 645791.109)
Band 1 Block=4299x1 Type=Byte, ColorInterp=Red
Band 2 Block=4299x1 Type=Byte, ColorInterp=Green
Band 3 Block=4299x1 Type=Byte, ColorInterp=Blue

Get information about all rasters in a table
Driver: GeoRaster/Oracle Spatial GeoRaster
Files: none associated
Size is 512, 512
Coordinate System is `'
Subdatasets:
SUBDATASET_1_NAME=geor:scott/tiger@,US_RASTERS_RDT_01,115
SUBDATASET_1_DESC=[4299x4299x3] CellDepth=8BIT_U SRID=26943
Corner Coordinates:
Upper Left ( 0.0, 0.0)
Lower Left ( 0.0, 512.0)
Upper Right ( 512.0, 0.0)
Lower Right ( 512.0, 512.0)
Center ( 256.0, 256.0)
gdalinfo georaster:scott/tiger,,us_rasters,georaster

Get information about a database raster
gdalinfo georaster:scott/tiger@spatialdb,us_rasters,georaster,georid=1
Driver: GeoRaster/Oracle Spatial GeoRaster
Files: none associated
Size is 4299, 4299
Coordinate System is:
PROJCS["NAD83 / California zone 3", …]
Origin = (1828467.919737499900000,646446.278891499970000)
Pixel Size = (0.304800999999999,-0.304800999999999)
Image Structure Metadata:
INTERLEAVE=PIXEL
COMPRESSION=NONE
Corner Coordinates:
Upper Left ( 1828467.920, 646446.279)
Lower Left ( 1828467.920, 645135.939)
Upper Right ( 1829778.259, 646446.279)
Lower Right ( 1829778.259, 645135.939)
Center ( 1829123.089, 645791.109)
Band 1 Block=512x512 Type=Byte, ColorInterp=Red
Overviews: 2149x2149, 1074x1074, 537x537, 268x268, 134x134
Band 2 Block=512x512 Type=Byte, ColorInterp=Green
Overviews: 2149x2149, 1074x1074, 537x537, 268x268, 134x134
Band 3 Block=512x512 Type=Byte, ColorInterp=Blue
Overviews: 2149x2149, 1074x1074, 537x537, 268x268, 134x134

Other Possibilities
• Convert a raster file from BSQ to BIP
• Re-project a raster file
• Get details about a file
• Compress a file
• Change format
gdal_translate -of GTiff -co "INTERLEAVE=PIXEL" sf1.bsq.tif sf1.tif
gdalwarp -s_srs epsg:26943 -t_srs epsg:32610 sf1.tif sf1_utm.tif
gdalinfo sf1.tif
gdal_translate -of gtiff sf1.tif sf1.cmp.tif -co compress=jpeg
gdal_translate -of jpeg sf1.tif sf1.jpg

Debugging Problems
• Get a trace of everything GDAL does
• For Oracle, will list all SQL calls
• Can write the trace to a file
--debug on
--debug on 2>gdal_trace.log

Exporting using GDAL
gdal_translate -of jpeg
georaster:scott/tiger@orcl111,us_rasters,georaster,georid=1
sf1.jpg
-co worldfile=YES
-of jpeg Use JPEG output
georaster:scott/tiger@orcl111 Database connection
us_rasters,georaster Destination table and column
georid=1 WHERE clause to select the raster to export
sf1.jpg Name of output JPEG file
-co worldfile=YES Produce a world file with georeferencing information

Creating a Spatial Index
• A spatial index is needed on the raster footprints
• Allows mapping tools to select the rasters to render for a map request.
insert into user_sdo_geom_metadata values (
'US_RASTERS',
'GEORASTER.SPATIALEXTENT',
sdo_dim_array(
sdo_dim_element('Easting', 1000000, 2000000, 1),
sdo_dim_element('Northing', 500000, 800000, 1)
),
40978);
create index us_rasters_sx
on us_rasters (georaster.spatialextent)
indextype is mdsys.spatial_index;

Georaster Tool

GeoRaster Tool: Viewer and ETL and more
• Single integrated tool
– ETL and Viewer
• Virtual Mosaic display is significantly enhanced
• New GUI to automate many SQL tasks:
– Creating GeoRaster tables
– Generating pyramids
– Rectifying images
– …
ORACLE CONFIDENTIAL –
118

How to Use it ?
• In your Oracle installation:
– $ORACLE_HOME/md/demo/georaster/tool
– Need to install the examples kit
• Requires a java environment
• Run the startup script for your platform:
– startGeoRasterTool.csh (Linux)
– startGeoRasterTool.bat (Windows)
• See the README.TXT file

GeoRaster ETL
• Simplify the process of loading large number of rasters
• Graphical user interface to select files to load and choose loading options
– Block size, compression, pyramiding, …
– Configures the load process in an XML description file
• Uses GDAL to perform the actual loading
– Includes an Oracle-built distribution of GDAL for popular platforms
– Windows 32 and 64-bit, Linux 64-bit
• Loads from all GDAL supported file formats
• Drives the loading in parallel
– Multiple concurrent GDAL jobs

Connect to the Database

Configure a Raster Import operation

Variable Column Values
• Use substitution variables in column values
• Will be replaced with actual values
• %FILENAME% is replaced with the name of the input raster file (without
directory)
• %n% (where n is a number) is replaced with a number that increments
sequentially from n
– For example %1001% will generate numbers 1001, 1002, 1003, etc.
– Use it to automatically number sequential identifiers (such as GEORID)

Save the Configuration File

Perform the Load

Configuration Files
<Batch>
<Command>
<Source>/media/sf_Spatial-Workshop/data/11-georaster/tiff/sf1.tif</Source>
<Format>GEORASTER</Format>
<Target>SCOTT.US_RASTERS,GEORASTER</Target>
<CreationOptionList>
<INSERT>(SOURCE_FILE,GEORASTER,GEORID) VALUES
('sf1.tif',sdo_geor.init('US_RASTERS_RDT_01'))</INSERT>
<GENPYRAMID>NN</GENPYRAMID>
<BLOCKYSIZE>512</BLOCKYSIZE>
<BLOCKING>OPTIMALPADDING</BLOCKING>
<BLOCKXSIZE>512</BLOCKXSIZE>
<BLOCKBSIZE>3</BLOCKBSIZE>
<SPATIALEXTENT>TRUE</SPATIALEXTENT>
</CreationOptionList>
</Command>
...
<Command> ... </Command>
</Batch>

Load Existing Configuration Files

Configuring and Debugging
• GDAL has a number of configuration options
• Set them via environment variables before launching the ETL
• GDAL_CACHEMAX = set memory cache
• CPL_DEBUG = trace all interactions
• See http://trac.osgeo.org/gdal/wiki/ConfigOptions
export GDAL_CACHEMAX=1024
./startGeoRasterETL.sh

What is Raster Algebra ?
• Raster Algebra (Map Algebra) is a set-based algebra for manipulating
geographic raster data, proposed by Dr. Dana Tomlin in the early 1980s. It is
widely used for cartographic modeling and spatial analysis.
• It includes a set of primitive algebraic operators applied on one or more
raster layers of similar dimensions to produce one or more new raster
layers or other values.
• There are 4 types of operations: local, focal, zonal and global.
• It provides a procedural or scripting language enabling very complex
operations
• However, all implementations are different.

Local Raster Algebra Operations
Local raster algebra
applies the same
operation point-by-point
for each cell location of
the input raster layers.
Z = (X+500)/(Y+100)
where X, Y are input
raster layers.
Z is the output layer
(Local operations use point-by-point processing of map values that occur at each map location. Source: Joseph K. Berry, Beyond Mapping III)

Raster Algebra and Analytics Engine
• The Raster Algebra Language is provided as an extension to PL/SQL
• Supports local raster algebra only in this release
• It includes an expression language for raster algebra operators
– general arithmetic, casting, logical and relational operators
• It includes 4 functions of Map Algebra:
– Cell-value based conditional queries: searches/masks cells based on boolean expression
– Cell-value based updates: update cells of a raster-based on boolean expression
– Arithmetic operations on cell values: ADD, DIVIDE, LOG, etc.
– Classification for raster segmentation: applies arithmetic expression to cells and then segments the
raster
• Polygon clipping based statistics generation functions, to support interactive analysis
on-the-fly

• PL/SQL provides
✓declaration of variables and constants
✓general expressions and functions
✓statements
✓Programs
• GeoRaster provides
✓raster algebra operators/expressions
✓raster algebra functions
The Raster Algebra Language
An extension to the PL/SQL language

Searching For Cells: A Sample Application of Raster Algebra
Select all cells for which the value of band 0 is between 10 and 50, band 1 is between 100 and 150 and band 2 is
between 200 and 245
declare
geor1 SDO_GEORASTER; -- source 3-band image
geor2 SDO_GEORASTER; -- result 3-band image
begin
select georaster into geor1 from georaster_table where georid = 1;
select georaster into geor2 from georaster_table where georid = 2;
sdo_geor_ra.findcells (
geor1,
' ({0}>=10&{0}<50)&({1}>=100&{1}<150)&({2}>200&{2}<245) ',
null, geor2, null, 'false','parallel=4'
);
update georaster_table set georaster = geor2 where georid = 2;
end;

The Raster Algebra Expressions
• Elements are variables, constants, operators (boolean, comparison or
arithmetic), and functions
• Variables refer to the values of cells in the raster(s) to process
– Form is {r,b} or {b}
– {2,0} refers to band 0 of the third input raster (for functions that operate on multiple
rasters)
– {1} refers to band 1 in the only input raster
• Constants are any number (integer or decimal)
• 0, 255, 0.44556,-273.15 are all valid numbers

• Arithmetic Operators are what you expect:
– +, -, *, /
• Comparison Operators:
– >, <, =, !=, <=, >=
• Boolean Operators:
– &, |, !
• Usual precedence rules for operators apply
– Use parentheses to force the order
• Boolean expressions return 0 or 1

• Functions:
– abs, sqrt, exp, log, ln, sin, cos, tan, sinh, cosh, tanh, arcsin, arccos, arctan, ceil, floor
– Apply to any arithmetic expression
• Casting a cell value to a specific data type:
– Useful when converting between different cell depths
– For example from 32BIT_FLOAT to 8BIT_UNSIGNED
– castint, castonebit, casttwobit, castfourbit, casteightbit
– Apply to any arithmetic expression

• An expression applies to all cells in the input rasters
• Input rasters must be coherent:
– Same row and column dimensions,
– Same geographical area.
• Can have different cell depth and different number of bands

• For example: ({0}+{1}+{2})/3 applies to all cells in the input raster.
– It computes the average of the values in bands 0, 1 and 2 for each cell.
• Same for boolean expressions: {0,1}>{1,1}*1.10 applies the comparison to
all cells of the two input rasters.
– It returns “true” when the value in band 1 in the first raster is more than 10% larger
than the same value in the second raster
– (actually returns 1 for “true” and 0 for “false”)

arithmeticUnaryOp:
+
| -
booleanBinaryOp:
&
| |
booleanUnaryOp:
!
rangeType:
castint
| castonebit
| casttwobit
| castfourbit
| casteightbit
numericFunction:
abs
| sqrt
| exp
| log
| ln
| sin
| cos
| tan
| sinh
| cosh
| tanh
| arcsin
| arccos
| arctan
| ceil
| floor
constantNumber:
double number
identifier:
{ID,band}
| {band}
ID:
integer number
band:
integer number

The Raster Algebra Functions
sdo_geor_ra.findCells Selects cells based on a boolean expression
sdo_geor_ra.rasterUpdate Updates cells of a raster based on a boolean expression
sdo_geor_ra.rasterMathOp Performs arithmetic operations
sdo_geor_ra.classify Applies arithmetic operations to cells and then segments
the raster
sdo_geor_ra.diff Computes the difference between two rasters
sdo_geor_ra.over Overlays two rasters
sdo_geor_ra.stack Computes a single-layer raster with statistics from
multiple input rasters and layers

Raster Algebra Functions
• Can use any number of GeoRaster objects in Raster Algebra
• Support sub-cell statistic mean (average) analysis
– Area weighted average based on intersection of irregular query boundary with cells
– For DEM, precision farming, climate modeling, low-resolution image analysis etc.
• All statistic generation functions are parallelized
– min, max, mean, median, mode, std
146

findCells: Cell Value-Based Conditional Queries
• Select all cells from the input raster that match the boolean condition
• Copy them into the output raster
• Other cells in the output raster are set to a chosen fixed value
• The output raster has the same structure as the input raster (same bands, same cell depth)
findCells
Input Raster Output Raster
Condition

findCells: example
• Input is a 3-band RGB color image
• Find all cells from the input raster where green dominates
– Value of band 1 is greater than value of band 0 and value of band 2
• Other cells set to white
sdo_geor_ra.findCells (
inGeoraster => gr1,
condition => '{1}>={0}&{1}>={2}',
storageParam => null,
outGeoraster => gr2,
bgValues => sdo_number_array (255,255,255)
);

rasterUpdate: Cell Value-Based Conditional Updates (Edits)
• There can be one or more input conditions
• Each condition is associated with a set of numeric values
• Select the cells from the input raster that match each of the boolean conditions
• When matched, replace its value(s) from the corresponding value(s) provided
• Values can be constants or RA expressions!
rasterUpdate
Input / Output Raster
Values
Conditions
a, b, c, d, e
a, b, c, d, e

rasterUpdate: example
• Replace cells with their dominant color
– e.g. when red dominates, set to red
sdo_geor_ra.rasterUpdate (
georaster => gr,
pyramidLevel => null,
conditions => sdo_string2_array(
'({0}>{1})&({0}>{2})', -- Red dominates
'({1}>{0})&({1}>{2})', -- Green dominates
'({2}>{0})&({2}>{1})’ -- Blue dominates
),
vals => sdo_string2_arrayset(
sdo_string2_array('255','0','0'),
sdo_string2_array('0','255','0'),
sdo_string2_array('0','0','255')
)
);
Those could also be
RA expressions

rasterMathOp: Mathematical Operations
• There are one or several input operations (arithmetic expressions)
• There are one or several input rasters and one output raster
• Apply each operation to the input raster(s)
• Write the result into one of the bands of the output raster
• The output raster has one band per operation
rasterMathOp
Input Rasters
Output Raster
Operations

rasterMathOp: example
• Create new single band raster in grayscale
• Compute new cells as the average of red, green and blue pixels
sdo_geor_ra.rasterMathOp (
inGeoraster => gr1,
operation => sdo_string2_array('({0}+{1}+{2})/3'),
storageParam => null
);

classify: Classification Operations
• Apply the expression on the input rasters
• Lookup the result in the Range Array
• Write the corresponding value from the Value Array in the output raster.
classify
Input Rasters
Output Raster
Expression
Range Array Value Array

classify: example
• Classify cells in four groups based on the average of the red, green and blue
pixels
• Output is a single-band raster with a cell depth of 2 bits.
sdo_geor_ra.classify (
inGeoraster => gr1,
expression => '({0}+{1}+{2})/3',
rangeArray => SDO_NUMBER_ARRAY(63,127,191),
valueArray => SDO_NUMBER_ARRAY(0,1,2,3),
storageParam => 'cellDepth=2BIT'
);
From To Group
0 63 0
64 127 1
128 191 2
192 255 3
333333330000123322310
333333332211233222122
333331122112332221212
333311201100112221230
333000332211233233312
333311231121111222123
333333302211233220022
333333002211233222121
222122100221123322212

diff: example
DIFF
Original Image Tree-removed Image Tree-only Image

A more complex example:
Historical Temperature Analysis
• A collection of 33 years of global temperature data for each month. In total, 396 layers stored in one
georaster object:
Year 1:
bands 12 to 23
Year 33:
bands 384 to 395
Year 1:
bands 0 to 11
…

Historical Temperature Analysis
1. The temperatures are in Kelvin:
convert them to Fahrenheit
2. Generate the Average (mean)
temperature for each month
3. Compute the Mean Absolute
Deviation per Month (the mean
difference from the average
across 33 years for each month).
'({0}-273.15)*9/5+32'
'({0}+{12}+{24}+{36}+{48}+{60}+{72}+{84}+{96}+{108}+{120}+{132}+
{144}+{156}+{168}+{180}+{192}+{204}+{216}+{228}+{240}+{252}+{264
}+{276}+{288}+{300}+{312}+{324}+{336}+{348}+{360}+{372}+{384})/3
3'
'(abs({1,0}-{0,0})+abs({1,0}-{0,12})+abs({1,0}-
{0,24})+abs({1,0}-{0,36})+abs({1,0}-{0,48})+abs({1,0}-
{0,60})+abs({1,0}-{0,72})+abs({1,0}-{0,84})+abs({1,0}-
{0,96})+abs({1,0}-{0,108})+abs({1,0}-{0,120})+abs({1,0}-
{0,132})+abs({1,0}-{0,144})+abs({1,0}-{0,156})+abs({1,0}-
{0,168})+abs({1,0}-{0,180})+abs({1,0}-{0,192})+abs({1,0}-
{0,204})+abs({1,0}-{0,216})+abs({1,0}-{0,228})+abs({1,0}-
{0,240})+abs({1,0}-{0,252})+abs({1,0}-{0,264})+abs({1,0}-
{0,276})+abs({1,0}-{0,288})+abs({1,0}-{0,300})+abs({1,0}-
{0,312})+abs({1,0}-{0,324})+abs({1,0}-{0,336})+abs({1,0}-
{0,348})+abs({1,0}-{0,360})+abs({1,0}-{0,372})+abs({1,0}-
{0,384}))/33'

Use PL/SQL to generate expressions
• Generate raster algebra
operations to convert all
bands from Kelvin to
Fahrenheit
• One operation per band
• Each operation looks like this:
– {b}-273.15)*9/5+32
– where b is the band number,
from 0 to 395
operation.Extend(numBands);
for b in 1 .. numBands loop
operation(i) := '({' || (b-1) || '}-273.15)*9/5+32';
end loop;
SDO_GEOR_RA.rasterMathOp(
inGeoRaster => gr1,
operation => operation,
storageParam => null,
nodata => 'TRUE',
nodataValue => -9999
);
operation sdo_string2_array := sdo_string2_array();

Historical Temperature Analysis: Average
June Average Temperature over 33 Years
159
(Source Temperature Data Courtesy of Remote Sensing Systems - http://www.remss.com/)

Historical Temperature Analysis: Average
December Average Temperature over 33 Years
160

Historical Temperature Analysis: Anomaly
June Temperature Mean Absolute Deviation Analysis
161

Historical Temperature Analysis: Anomaly
December Temperature Mean Absolute Deviation Analysis

NDVI Computation
• Normalized Difference Vegetation Index
• A widely used index, enabling users to quickly identify
vegetated areas and monitor the growth and condition
of plants.
• Calculated from the visible and near-infrared light
reflected by vegetation.
NDVI = (NIR — VIS)/(NIR + VIS)
• Values range from -1.0 to 1.0
Images courtesy of NASA Earth Observation
NDVI Interpretation
~ -1.0 Water
~ 0.0 Barren areas
~ 0.2-0.4 Shrub/Grass
~ 1.0 Forest

NVDI
image
NDVI Computation
• For Landsat Thematic Mapper (TM) imagery,
the standard NDVI computation formula is:
(TM4 - TM3) / (TM4 + TM3).
• Implement using Oracle raster algebra:
ETM+543
false color
image
Images courtesy of PCI Geomatics
inGeoraster => gr1,
operation =>
sdo_string2_array(
'({3}-{2})/({3}+{2})'),
storageParam => 'celldepth=32bit_real'
);

Tasseled Cap Transformation (TCT)
• Bands of an image converted into a new
set of bands with defined interpretations
that are useful for vegetation mapping.
• Each tasseled-cap band is created by the
sum of image band 1 * a coefficient +
image band 2 * a coefficient etc.
• Coefficients are derived statistically from
images and empirical observations.
Source TM 123 Color Image (upper left), Brightness (upper
right), Greenness (lower left) and Wetness (lower right).
Landsat 5 TM Image Courtesy of the U.S. Geological Survey

Tasseled Cap Transformation Example
inGeoraster => gr1,
operation => sdo_string2_array(
'0.3561*{0}+0.3972*{1}+0.3904*{2}+0.6966*{3}+0.2286*{4}+0.1596*{6}',
'(-0.3344)*{0}-0.3544*{1}-0.4556*{2}+0.6966*{3}-0.0242*{4}-0.2630*{6}',
'0.2626*{0}+0.2141*{1}+0.0926*{2}+0.0656*{3}-0.7629*{4}-0.5388*{6}',
'0.0805*{0}-0.0498*{1}+0.1950*{2}-0.1327*{3}+0.5752*{4}-0.7775*{6}',
'(-0.7252)*{0}-0.0202*{1}+0.6683*{2}+0.0631*{3}-0.1494*{4}-0.0274*{6}',
'0.4000*{0}-0.8172*{1}+0.3832*{2}+0.0602*{3}-0.1095*{4}+0.0985*{6}’
),
storageParam => 'celldepth=32bit_real'
);

Cartographic Modeling
• Assume that a hypothetical
cartographic model involves 7
different raster layers (stored in 7
georasters) and has an expression as
follows:
if (100 < layer1 <= 500) &
(layer2 == 3 or layer2 == 10) &
((((layer3+layer4) * log(layer5) / sqrt(layer5)) >= layer6) or (layer7 != 1) )
then
output = 1
else
output = 0
expr :=
'((100<{0,0})&({0,0}<=500))&(({1,0}=3)
|({1,0}=10))&(((({2,0}+{3,0})*log({4,0
})/sqrt({4,0}))>={5,0})|({6,0}!=1))’;
inGeoRasters => my_cursor,
operation =>
sdo_string2_array(expr),
storageParam => 'celldepth=1bit'
nodata => 'TRUE',
nodataValue => 0,
parallelParam => ’parallel=4'
);

Image Processing

Image Processing PL/SQL Packages
• SDO_GEOR
– Subsetting, reprojection, rectification, warping, scaling, affine transformation,
merging, statistics …
• SDO_GEOR_AGGR
– Mosaicing, appending, virtual mosaic …
• SDO_GEOR_IP
– Filtering, normalization, dodging, stretching, equalization, histogram matching …
• SDO_GEOR_GDAL
– GDAL image processing functions
169

Package SDO_GEOR_AGGR
170
mosaicSubset() Advanced Large-Scale Image Mosaicking
append() Large-Scale Image Appending
getMosaicSubset() Virtual Mosaic

Large-Scale Mosaicking: sdo_geor_aggr.mosaicSubset()
✓Parallelized and scalable
✓Source images can be tables, views or a SQL statement
✓Source images can be rectified or unrectified or in different CS
✓Can do internal resampling, reprojection or rectification
✓Allows gaps, no data, and overlapping regions
✓Provides 8 common point rules (max, min, avg, LATEST, OLDEST ...)
✓User defined priority for overlapping regions (Date or SQL ORDER BY)
✓Simple color balancing (linear stretch and normalization)

An Example – Mosaic of Landsat Images
Source images:
✓ 31 Landsat 5 Level 1T TM images, from the U.S.
Geological Survey.
✓ The total size of the image set is 11.9 GB.
✓ 15 images are in UTM zone 11N
✓ 16 images are in UTM zone 10N projection.
✓They overlap each other.
Output mosaic:
about 7.7GB in size

Large-Scale Appending: sdo_geor_aggr.append()
• Used to append or update a large physical mosaic using newly acquired
smaller images
• The source image can be on ANY SIDE of the target image!
• Same functionality as mosaicking:
– Source images can be rectified or unrectified or in different CS
– Support internal resampling, reprojection or rectification
– Supports gaps, no data, and overlapping regions
– Support 8 common point rules (max, min, avg, LATEST, OLDEST, etc...)
– Simple color balancing (linear stretch and normalization)

Virtual Mosaic: sdo_geor_aggr.getMosaicSubset()
• A virtual mosaic is defined as any large collection of georeferenced
GeoRaster objects (images) that is treated as if it is a single GeoRaster
object (physical mosaic).
• Three ways to define a virtual mosaic:
– a list of GeoRaster tables
– a database view with a GeoRaster column
– a SQL query statement (i.e., a CURSOR)
• A virtual mosaic can contain unlimited number of images of any size
• There is no need to define a description file for the virtual mosaic

Virtual Mosaic
• To query or process the virtual mosaic, use:
– sdo_geor_aggr.getMosaicSubset()
to perform on-the-fly cropping queries over the virtual mosaic
– sdo_geor_aggr.mosaicSubset()
to perform a query and store the mosaicked subset as a GeoRaster object
• To facilitate application development, use:
– sdo_geor_aggr.validateForMosaicSubset()
– sdo_geor_aggr.getMosaicExtent()
– sdo_geor_aggr.getMosaicResolution()

Virtual Mosaic
Same functionality as large-scale mosaicking
• Source images can be rectified or unrectified or in different CS
• Support internal resampling, reprojection or rectification
• Supports gaps, no data, and overlapping regions
• Support 8 common point rules (max, min, avg, LATEST, OLDEST, etc...)
• User defined priority for overlapping regions (Date or SQL ORDER BY)
• Simple color balancing (linear stretch and normalization)

An Example - Virtual Mosaic
Overview Zoom In Zoom In Further
(Image Courtesy of the U.S. Geological Survey)

Virtual Mosaic Use Cases
✓DOQs can be stored as is - but used as seamless mosaic (as fast as physical
mosaics)
✓Users may not want to mosaic DEM’s
✓Store large volume of imagery without making too many extra copies (save
disk spaces)
✓New images coming in and immediately displayed on the mosaic (dynamic
updates)
✓The same images can be displayed or removed in different virtual mosaics
(flexible model)

Color Balancing Methods for Mosaicking
• Linear Stretching
– (1) to the given min and max value or (2) to the given reference image band by band
• Statistic Matching (Normalization)
– (1) to the given mean and std value or
– (2) to the given reference image band by band or
– (3) to use overlapping areas image by image and band by band
• Histogram Matching
– (1) to the given histogram or
179

Example – Image Mosaic with Statistic Matching
180

181

182

183

Package SDO_GEOR
184
subset() Subsetting
reproject() Reprojection
rectify() Rectification
scale() Scaling
merge() Merging
warp() Warping
dodge(): Affine transformation
generateStatistics() Statistics

Image Rectification and Orthorectification
• Rectification of georeferenced raw images
• Orthorectification of georeferenced raw image using DEM
• Both support on-the-fly cropping queries or for persistent storage
rectify
(Image Courtesy of Digital Globe)

Package SDO_GEOR_IP
186
stretch() Linear stretching
piecewiseStretch() Piecewise stretching
equalize() Equalization
filter() Filtering
histogramMatch() Histogram matching
normalize() Normalization
dodge(): Dodging

Example – Image Dodging
187
This dodging operation uses an adaptive image enhancement
method to make the image tone more balanced, that is, the
darker area becomes brighter and the bright area becomes
darker.

188

Example – High Pass Filtering
189
Applying a high-pass filter on an image can enhance the
details of the image or even detect the edges inside the
image.

190

Package SDO_GEOR_GDAL
• Integrates GDAL into Oracle database server
• Provides server-side raster data loading, exporting, and in-database terrain
analysis functions
• Integrated as an "extproc" library
• Enables any Raster analysis function available in GDAL with the Oracle DB
191
dem() GDAL_DEM: process a Digital Elevation Model (DEM) into a new raster (color relief,
hill shading, slope …)
info() GDALINFO: returns information from a raster
translate() GDAL_TRANSLATE: Converts raster data between different formats; perform
operations like subsetting, resampling, rescaling pixels …

Managing Raster Structures

Dropping Raster Tables
• Cannot drop a RDT if it contains blocks for existing rasters:
• First delete the objects, truncate the raster table or drop it
drop table us_rasters_rdt_01
*
ERROR at line 1:
ORA-00604: error occurred at recursive SQL level 1
ORA-13391: The referenced raster data table(RDT) cannot be dropped.
ORA-06512: at "MDSYS.MD", line 1723
ORA-06512: at "MDSYS.MDERR", line 17
ORA-06512: at line 1
ORA-06512: at line 188
SQL> delete from us_rasters;
4 rows deleted.
SQL> drop table us_rasters_rdt_01;
Table dropped.

Raster Metadata
• Check views USER_SDO_GEOR_SYSDATA or ALL_SDO_GEOR_SYSDATA
• Lists all the rasters in your schema or in all schemas
Name Null? Type
--------------------- -------- ----------------------
OWNER NOT NULL VARCHAR2(32)
TABLE_NAME NOT NULL VARCHAR2(32)
COLUMN_NAME NOT NULL VARCHAR2(1024)
METADATA_COLUMN_NAME VARCHAR2(1024)
RDT_TABLE_NAME NOT NULL VARCHAR2(32)
RASTER_ID NOT NULL NUMBER
OTHER_TABLE_NAMES MDSYS.SDO_STRING_ARRAY

Checking for Errors
This function checks for the following errors:
• The RDT name is not unique.
• The GeoRaster table does not exist.
• The GeoRaster column does not exist.
• The GeoRaster object does not exist.
• The GeoRaster object is non-empty or nonblank, but the RDT does not exist.
• Duplicate GeoRaster objects exist (that is, one or more non-unique combinations of RDT
and raster ID)
select * from table( SDO_GEOR_ADMIN.checkSysdataEntries );

Fixing Errors
• Fixes errors reported by SDO_GEOR_ADMIN.checkSysdataEntries()
• Check for any « dangling » raster blocks
– Blocks in an RDT that are not used by any georaster object
DECLARE
corrected_errors SDO_STRING2_ARRAY;
BEGIN
corrected_errors:= sdo_geor_admin.MaintainSysdataEntries;
for i in 1..corrected_errors.count loop
dbms_output.put_line(corrected_errors(i));
end loop;
END;
/

Checking for Errors
• Check for any « dangling » RDTs
– RDT not used by any georaster object
• Check for any « dangling » raster blocks
– Blocks in an RDT that are not used by any georaster object
select * from table(SDO_GEOR_ADMIN.listUnregisteredRDT);
select * from table( SDO_GEOR_ADMIN.listDanglingRasterData );

Solving RDT name conflicts
• Names of RDT tables must be globally unique
• Cannot have two RDTs with the same name in different schemas
• To rename an RDT:
• This renames the table and updates all georaster objects.
EXECUTE sdo_geor_utl.renameRDT('US_RASTERS_RDT_01', 'US_RASTERS_RDT_02');

Moving Rasters between Databases
• Use export and import tools
– EXP/IMP or EXPDP/IMPDP
• Export the raster table(s) as well as the RDT(s).
• Import into new database or schema
expdp scott/tiger directory=DATA_PUMP_DIR
dumpfile=us_rasters.dmp
tables=(us_rasters, us_rasters_rdt_01)
impdp scott/tiger directory=DATA_PUMP_DIR
dumpfile=us_rasters.dmp
tables=(us_rasters, us_rasters_rdt_01)
exclude=trigger:"like 'GRDMLTR%'"
Exclude the DML triggers automatically created by Georaster (or just ignore the creation errors)

Moving Rasters between Databases
• Use transportable tablespaces
– EXPDP/IMPDP
• « Export » the tablespace(s)
• « Import » into new schema
• Register the raster objects
expdp system/manager directory=DATA_PUMP_DIR
dumpfile=us_rasters_ts.dmp
transport_tablespaces=us_rasters_ts
impdp system/manager directory=DATA_PUMP_DIR
dumpfile=us_rasters_ts.dmp
transport_datafiles=(us_rasters_ts01.dbf,us_rasters_ts02.dbf)
EXECUTE sdo_geor_admin.registerGeoRasterObjects;

Java API

• Provides helper functions for
Georaster access in java
• Package oracle.spatial.georaster
• Fully functional examples with
source code
– Loader
– Exporter
– Viewer
• Classes:
– JGeoRaster: Maps to the
SDO_GEORASTER type
– JGeoRasterMeta: Deals with raster
metadata
– Jraster: Retrieve raster data
– GeoRasterImage: Retrieve raster for
visualization
– SdoGeorPKG: Access PL/SQL APIs in
package SDO_GEOR
Java API

Fetching Rasters
// Construct SQL query
String sqlQuery = "SELECT GEORASTER FROM US_RASTERS WHERE GEORID=1"
// Execute query
Statement stmt = dbConnection.createStatement();
OracleResultSet rs = (OracleResultSet)stmt.executeQuery(sqlQuery);
// Fetch results
while (rs.next()) {
// Extract JDBC object from record into structure
STRUCT dbObject = (STRUCT) rs.getObject(1);
// Import from structure into Geometry object
JGeoRaster geor = new JGeoRaster (dbObject);
}

Fetching Raster Data
// Get JRaster object to fetch raster content
JRaster jr = geor.getRasterObject();
// Get one raster block
Byte[] rb = jr.getRasterBlock (
pyramidLevel, bandBlockNumber, rowBlockNumber,
columnBlockNumber);
// Get entire raster data
BLOB rd = jr.getRasterData (pyramidLevel);
columnBlockNumber);
// Get a subset of a raster
Byte[] rsub = jr.getRasterSubset (
pyramidLevel, window, layerNo, cellDepth, compression);

What´s New with Oracle Database 19c
The following slides (until the end) are contributed by Hans Viehmann, Oracle
Corporation

Native JPEG 2000 Compression!
✓The same as JP2 file. Stored in a single BLOB. Image can be tiled
✓High compression ratio. For the same quality, 2-7 times smaller than JPEG
image
✓Support both lossless and lossy compressions
✓No separate pyramids. Pyramid level is limited by tile dimension sizes
✓Support 8bit and 16bit integer images
✓No limit on number of bands
✓Support very large size image. Size is limited by memory and max number
of tiles (max_mem_size / 20 * 65535)

Using JP2-compressed Rasters
• New procedures:
– sdo_geor.compressJP2()
– sdo_geor.decompressJP2()
• All key functions read JP2-compressed images:
– getRasterSubset, subset, rectify, large scale mosaicking, virtual mosaic, and raster
algebra.
• Most functions, such as getRasterSubset, subset, and virtual mosaic
support partial decompressing (faster).
• A few require full image decompressing (slower).

Importing JP2 rasters
• Direct loading with GDAL without decompression and recompression!
• Content of input loaded as-is into the raster table
• Complete with pyramid!
gdal_translate -of georaster /images/image_3.jp2
georaster:georaster/georaster@my_db,image_table,raster
-co "insert=(id,label,raster) values (3, 'image_3', sdo_geor.init('rdt_table', 3)"
-co compress=jp2-f

GDAL support in GeoRaster
• A new SDO_GEOR_GDAL PL/SQL package is added
– integrates GDAL into Oracle database server
– It provides server-side raster data loading, exporting, and in-database terrain analysis
functions
– Integrated as an extproc library
• Enables any Raster analysis function available in GDAL with the Oracle DB

• Linear stretching
• Piecewise stretching
• Equalization
• Normalization
• Histogram matching
• Dodging
• Filtering
• Warping
• Affine transformation
New Image Processing: Package SDO_GEOR_IP

This dodging operation uses an adaptive image enhancement
method to make the image tone more balanced, that is, the
darker area becomes brighter and the bright area becomes
darker.

Applying a high-pass filter on an image can enhance the
details of the image or even detect the edges inside the
image.

Color Balancing Methods for Mosaicking
• Linear Stretching
– (1) to the given min and max value or (2) to the given reference image band by band
• Statistic Matching (Normalization)
– (1) to the given mean and std value or
• Histogram Matching
– (1) to the given histogram or

Raster Algebra
• 26 New Operators added to the Raster Algebra language
– conditionalExpr (same as if … then … else …)
– ^ (i.e., XOR), % (i.e., MODULO), POWER, FACTORIAL
– Local statistic operators
– many others
• 3 New Overlay Functions
– DIFF
– OVER
– STACK

Example – DIFF Operation
DIFF
Original Image Tree-removed Image Tree-only Image

Raster Algebra
• Can use any number of GeoRaster objects in Raster Algebra
– Was limited to 8 raster objects
• Support sub-cell statistic mean (average) analysis
– Area weighted average based on intersection of irregular query boundary with cells
– For DEM, precision farming, climate modeling, low-resolution image analysis etc.
• Parallelized all statistic generation functions
– min, max, mean, median, mode, std

Core Georaster Functions (SDO_GEOR)
• New functions for cell-level access
– getCellValues, evaluateDoubles, changeCellvalues, getBitmapMaskValues
– Allow multiple points access in one function call
• getRasterSubset can return cell values in array directly
– Simplify development and improve query performance

Core Georaster Functions (SDO_GEOR)
• DEM based 3D surface area computation with parallelism (using 3D
triangles)
• generateBitmapPyramid (needed to run provided scripts)
• Area weighted sub-cell statistic average/mean analysis
• Color ramping functions to automatically generate colormaps for grids or
images
• Raster warping function

Example – Color Ramping

Example – generateBitmapPyramid
Old Pyramid
(with missing and dash lines)
New Pyramid
(no dash lines)
(Map Data Courtesy of G.ON)

Core Georaster Functions Performance
• Memory control functions
– Enable users to control how much memory to use
– Allows bigger memory usage
– Improve performance of all GeoRaster operations by limiting I/O overhead.
• More parallelism
– Rectification, compression, statistics generation
• Internal “sysdata” caching
– Improves performance of repeated calls of simple access functions on same raster
– Example: repeated calls of getCellValue and changeCellValue by about 30%.

Java API
• New Virtual Mosaic API
• Speeds up integration
• Used by MapViewer
– New Virtual Mosaic theme

Viewer and ETL
• Single integrated tool
– ETL and Viewer
• Virtual Mosaic display is significantly enhanced
• New GUI to automate some SQL tasks:
– Creating GeoRaster tables
– Generating pyramids
– Rectifying images
– …

GeoRaster 19c New Features
• Supports On-the-fly Raster Algebra
– All subprograms in the SDP_GEOR_RA package support putting the result data in temporary BLOB so that users
can do raster analysis and cartographic modeling on-the-fly to support various real time applications
– Supported in both PL/SQL API and JAVA API
• GeoRaster Must Be Enabled at Schema Level
– Previously GeoRaster is enabled for the whole database
– This enhances database security for the cloud

Temporal Statistics with Raster Algebra

Two Main Raster Statistical Functions
SDO_GEOR.GenerateStatistics and SDO_GEOR.Stack
• SDO_GEOR.GenerateStatistics
– Operates on a single raster
– Performs statistics across all the rows and columns of a raster, or a specified clip region
• SDO_GEOR_RA.Stack – New in Oracle 12.2
– Operates on a set of rasters. Great for Temporal Statistical Analysis.
– Dynamically “Stacks” the rasters, and you pick which bands to compute statistics on
– Rasters to process can be part of a temporal query, for example, a five day query:
• SELECT raster FROM rainfall_rasters WHERE timestamp BETWEEN timestamp_value AND timestamp_value + 5
• The timestamp query could return many rasters to dynamically “Stack” by SDO_GEOR_RA.Stack.
– Raster 1 contains rainfall at time t1
– Raster 2 contains rainfall at time t2
– And so on….
• Generate rainfall statistics across many rasters
– Can also apply a clip region

Two Main Raster Statistical Functions
SDO_GEOR.GenerateStatistics and SDO_GEOR.Stack - Results
• SDO_GEOR.GeneateStatistics results
– min=0
– max=65535
– mean=14093.9948936495
– median=3005
– mode=0
– std=23642.628261204
• SDO_GEOR_RA.Stack results
– Generates a new raster with the statistical results summarized for each (row,col)
– Result values can be interrogated very quickly with either SDO_GEOR.getCellValue
– Or, you can also run SDO_GEOR.GeneateStatistics for statistics of a Stack result

Topics
236
Concepts
Storage Model
Loading and exporting Rasters
Raster Viewer and ETL
Mosaicing and other Image Processing
What´s New with Oracle DB 19c
1
2
3
4
5
6
7

20190704_AGIT_Georaster_ImageryData_KPatenge

More Related Content

What's hot (20)

Similar to 20190704_AGIT_Georaster_ImageryData_KPatenge (20)

More from Karin Patenge (19)

Recently uploaded (20)

20190704_AGIT_Georaster_ImageryData_KPatenge