Generic Image Processing With Climb
1 like314 views
Climb is a generic image processing library written in Common Lisp that provides algorithms like thresholding, mathematical morphology, and watershed segmentation. It aims to be fully generic by handling different image formats, pixel types, and storage schemes. Climb uses morphers to allow algorithms to operate on different image types transparently and property descriptions to select algorithm implementations based on image structure details. The library also provides tools for chaining algorithms together into complex image processing pipelines.
![Generic Image Processing With Climb
Laurent Senta Christopher Chedeau Didier Verna
senta@lrde.epita.fr christopher.chedeau@lrde.epita.fr didier.verna@lrde.epita.fr
Epita Research and Development Laboratory
14-16 rue Voltaire, 94276 Le Kremlin-Bicêtre
Paris, France
ABSTRACT a dynamic language. Common Lisp provides the necessary
Categories and Subject Descriptors flexibility and extensibility to let us consider several alter-
I.4.0 [Image Processing And Computer Vision]: Gen- nate implementations of the model provided by Olena, from
eral—image processing software; D.2.11 [Software Engi- the dynamic perspective.
neering]: Software Architectures—Data abstraction, Domain-
specific architectures First, we provide a survey of the algorithms readily available
in Climb. Next, we present the tools available for compos-
ing basic algorithms into more complex Image Processing
General Terms chains. Finally we describe the generic building blocks used
Design, Languages to create new algorithms, hereby extending the library.
Keywords 2. FEATURES
Generic Image Processing, Common Lisp
Climb provides a basic image type with the necessary load-
ing and saving operations, based on the lisp-magick library.
1. INTRODUCTION This allows for direct manipulation of many images formats,
Climb is a generic image processing library written in Com- such as PNG and JPEG. The basic operations provided by
mon Lisp. It comes with a set of built-in algorithms such as Climb are loading an image, applying algorithms to it and
erosion, dilation and other mathematical morphology oper- saving it back. In this section, we provide a survey of the
ators, thresholding and image segmentation. algorithms already built in the library.
The idea behind genericity is to be able to write algorithms
only once, independently from the data types to which they 2.1 Histograms
will be applied. In the Image Processing domain, being fully A histogram is a 2D representation of image information.
generic means being independent from the image formats, The horizontal axis represents tonal values, from 0 to 255
pixels types, storage schemes (arrays, graphs, dimensions) in common grayscale images. For every tonal value, the
etc. Such level of genericity, however, may induce an impor- vertical axis counts the number of such pixels in the image.
tant performance cost. Some Image Processing algorithms may be improved when
the information provided by a histogram is known.
In order to reconcile genericity and performance, the LRDE1
has developed an Image Processing platform called Olena2 .
Olena is written in C++ and uses its templating system Histogram Equalization. An image (e.g. with a low con-
abundantly. Olena is a ten years old project, well-proven trast) does not necessarily contain the full spectrum of in-
both in terms of usability, performance and genericity. [2] tensity. In that case, its histogram will only occupy a nar-
row band on the horizontal axis. Concretely, this means
In this context, the goal of Climb is to use this legacy for that some level of detail may be hidden to the human eye.
proposing another vision of the same domain, only based on Histogram equalization identifies how the pixel values are
spread and transforms the image in order to use the full
1
EPITA Research and development laboratory range of gray, making details more apparent.
2
http://olena.lrde.epita.fr
Plateau Equalization. Plateau equalization is used to re-
veal the darkest details of an image, that would not oth-
erwise be visible. It is a histogram equalization algorithm
applied on a distribution of a clipped histogram. The pixels
values greater than a given threshold are set to this thresh-
old while the other ones are left untouched.
2.2 Threshold](https://image.slidesharecdn.com/senta-12-els-120502060510-phpapp01/85/Generic-Image-Processing-With-Climb-1-320.jpg)
![Thresholding is a basic binarization algorithm that converts Mathematical Morphology can also be applied on grayscale
a grayscale image to a binary one. This class of algorithms images, leading to news algorithms:
works by comparing each value from the original image to a
threshold. Values beneath the threshold are set to False in
the resulting image, while the others are set to True. TopHat Sharpen details
Laplacian operator Extract edges
Climb provides 3 threshold algorithms:
Hit-Or-Miss Detect specific patterns
Basic threshold. Apply the algorithm using a user-defined 3. COMPOSITION TOOLS
threshold on the whole image. Composing algorithms like the ones described in the previ-
ous section is interesting for producing more complex image
processing chains. Therefore, in addition to the built-in al-
Otsu Threshold. Automatically find the best threshold value gorithms that Climb already provides, a composition infras-
using the image histogram. The algorithm picks the value tructure is available.
that minimizes the spreads between the True and False do-
mains as described by Nobuyuki Otsu [4]. 3.1 Chaining Operators
3.1.1 The $ operator
The most essential form of composition is the sequential one:
Sauvola threshold. In images with a lot of variation, global algorithms are applied to an image, one after the other.
thresholding is not accurate anymore, so adaptive threshold-
ing must be used. The Sauvola thresholding algorithm [5] Programming a chain of algorithms by hand requires a lot
picks the best threshold value for each pixel from the given of variables for intermediate storage of temporary images.
image using its neighbors. This renders the code cumbersome to read an maintain.
2.3 Watershed To facilitate the writing of such a chain, we therefore pro-
vide a chaining operator which automates the plugging of
Watershed is another class of image segmentation algorithms.
an algorithm’s output to the input of the next one. Such
Used on a grayscale image seen as a topographic relief, they
an operator would be written as follows in traditional Lisp
extract the different basins within the image. Watershed al-
syntax:
gorithms produce a segmented image where each component
is labeled with a different tag. ( image−s a v e
( algo2
Climb provides a implementation of Meyer’s flooding algo- ( algo1
rithm [3]. Components are initialized at the “lowest” region ( image−l o a d ‘ ‘ image . png ’ ’ )
of the image (the darkest ones). Then, each component is param1 )
extended until it collide with another one, following the im- param2 )
age’s topography. ‘ ‘ output . png ’ ’ )
2.4 Mathematical Morphology Unfortunately, this syntax is not optimal. The algorithms
Mathematical Morphology used heavily in Image Processing. must be read in reverse order, and arguments end up far
It defines a set of operators that can be cumulated in order away from the function to which they are passed.
to extract specific details from images.
Inspired by Clojure’s Trush operator (->) 3 and the JQuery4
Mathematical Morphology algorithms are based on the ap- chaining process, Climb provides the $ macro which allows
plication of a structuring element to every points of an im- to chain operations in a much clearer form.
age, gathering information using basic operators (e.g. logical $ ( ( image−l o a d ‘ ‘ image . png ’ ’ )
operators and, or) [6]. ( a l g o 1 param1 )
( a l g o 2 param2 )
( image−s a v e ) )
Erosion and Dilation. Applied on a binary image, these
algorithms respectively erodes and dilate the True areas in
the picture. Combining these operators leads to new algo- 3.1.2 Flow Control
rithms: In order to ease the writing of a processing chain, the $
macro is equipped with several other flow control operators.
Opening Apply an erosion then a dilation. This removes
the small details from the image and opens the True • The // operator splits the chain into several indepen-
shapes. dent subchains that may be executed in parallel.
3
http://clojure.github.com/clojure/clojure.
Closing Apply a dilation then an erosion. This closes the core-api.html#clojure.core/->
4
True shapes. http://api.jquery.com/jQuery/](https://image.slidesharecdn.com/senta-12-els-120502060510-phpapp01/85/Generic-Image-Processing-With-Climb-2-320.jpg)
![• The quote modifier (’) allows to execute a function Value morpher. A value morpher operates a dynamic trans-
outside the actual processing chain. This is useful for lation from one value type to another. For instance, an RGB
doing side effects (see the example below). image can be used as a grayscale one when seen through a
morpher that returns the pixels intensity instead of their
• The $1, $2, etc. variables store the intermediate results original color. It is therefore possible to apply a watershed
from the previous executions, allowing to use them ex- algorithm on a colored image in a transparent fashion, with-
plicitly as arguments in the subsequent calls in the out actually transforming the original image into a grayscale
chain. one.
• The # modifier may be used to disrupt the implicit
chaining and let the programmer use the $1, etc. vari-
ables explictly. Content morpher. A content morpher operates a dynamic
translation from one image structure to another. For in-
stance, a small image can be used as a bigger one when seen
These operators allow for powerful chaining schemes, as il- through a morpher that returns a default value (e.g. black),
lustrated below: when accessing coordinates outside the original image do-
( $ ( l o a d ‘ ‘ l e n a g r a y . png ’ ’ ) main. It is therefore possible to apply an algorithm built
for images with a specific size (e.g. power of two), without
(// having to extend the original image.
( ’ ( timer −s t a r t )
( otsu ) Possible uses for content morphers include restriction (parts
’ ( timer −p r i n t ”Otsu ”) of the structures being ignored), addition (multiple struc-
( s a v e ” o t s u . png ”) ) ; $1 tures being combined) and reordering (structures order be-
ing modified).
( ’ ( timer −s t a r t )
( s a u v o l a ( box2d 1 ) ) 4. EXTENSIBILITY
’ ( timer −p r i n t ”Sauvol a 1 ”) The pixels of an image are usually represented aligned on
( s a v e ” s a u v o l a 1 . png ”) ) ; $2 a regular 2D grid. In this case, the term “pixel” can be
interpreted in two ways, the position and the value, which
( ’ ( timer −s t a r t ) we need to clearly separate. Besides, digital images are not
( s a u v o l a ( box2d 5 ) ) necessarily represented on a 2D grid. Values can be placed
’ ( timer −p r i n t ”Sauvol a 5 ”) on hexagonal grids, 3D grids, graphs etc. In order to be
( s a v e ” s a u v o l a 5 . png ”) ) ) ; $3 sufficiently generic, we hereby define an image as a function
from a position (a site) to a value: img(site) = value.
(//
(#( d i f f $1 $2 ) A truly generic algorithm should not depend on the under-
( save ‘ ‘ d i f f −otsu−s a u v o l a 1 . png ’ ’ ) ) lying structure of the image. Instead it should rely on higher
(#( d i f f $1 $3 ) level concepts such as neighborhoods, iterators, etc. More
( save ‘ ‘ d i f f −otsu−s a u v o l a 5 . png ’ ’ ) ) precisely, we have identified 3 different levels of genericity,
(#( d i f f $2 $3 ) as explained below.
( save ‘ ‘ d i f f −s a u v o l a 1 −s a u v o l a 5 . png ’ ’ ) ) ) )
4.1 Genericity on values
Genericity on values is based on the CLOS dispatch. Us-
3.2 Morphers ing multimethods, generic operators like comparison and
As we saw previously, many Image Processing algorithms
addition are built for each value type, such as RGB and
result in images of a different type than the original image
Grayscale.
(for instance the threshold of a grayscale image is a boolean
one). What’s more, several algorithms can only be applied
on images of a specific type (for instance, watershed may 4.2 Genericity on structures
only be applied to grayscale images). The notion of “site” provides an abstract description for any
kind of position within an image. For instance, a site might
Programming a chain of algorithms by hand requires a lot of be an n-uplet for an N-D image, a node within a graph, etc.
explicit conversion from one type to another. This renders
the code cumbersome to read an maintain. Climb also provides a way to represent the regions of an
image through the site-set object. This object is used in or-
To facilitate the writing of such a chain, we therefore provide der to browse a set of coordinates and becomes particularly
an extension of the Morpher concept introduced in Olena handy when dealing with neighborhoods. For instance, the
with SCOOP2 [1] generalized to any object with a defined set of nearest neighbors, for a given site, would be a list of
communication protocol. Morphers are wrappers created coordinates around a point for an N-D image, or the con-
around an object to modify how it is viewed from the outside nected nodes in a graph.
world.
4.3 Genericity on implementations
Two main categories of morphers are implemented: value Both of the kinds of genericity described previously allow
morphers and content morphers. algorithm implementors to produce a single program that](https://image.slidesharecdn.com/senta-12-els-120502060510-phpapp01/85/Generic-Image-Processing-With-Climb-3-320.jpg)
![works with any data type. Some algorithms, however, may Performance By improving the state of the current im-
be implemented in different ways when additional informa- plementation, exploring property-based algorithm op-
tion on the underlying image structure is available. For in- timization and using the compile-time facilities offered
stance, iterators may be implemented more efficiently when by the language.
we know image contents are stored in an array. We can also
distinguish the cases where a 2D image is stored as a matrix
or a 1D array of pixels etc. Climb is primarily meant to be an experimental plateform
for exploring various generic paradigms from the dynamic
Climb provides a property description and filtering system languages perspective. In the future however, we also hope
that gives a finer control over the algorithm selection. This is to reach a level of usability that will trigger interest amongst
done by assigning properties such as :dimension = :1D, :2D, Image Processing practitioners.
etc. to images. When defining functions with the defalgo
macro, these properties are handled by the dispatch system. 6. REFERENCES
The appropriate version of a function is selected at runtime, [1] Th. G´raud and R. Levillain. Semantics-driven
e
depending on the properties implemented by the processed genericity: A sequel to the static C++ object-oriented
image. programming paradigm (SCOOP 2). In Proceedings of
the 6th International Workshop on Multiparadigm
5. CONCLUSION Programming with Object-Oriented Languages
Climb is a generic Image Processing library written in Com- (MPOOL), Paphos, Cyprus, July 2008.
mon Lisp. It is currently architected as two layers targeting [2] R. Levillain, Th. G´raud, and L. Najman. Why and
e
two different kinds of users. how to design a generic and efficient image processing
framework: The case of the Milena library. In
Proceedings of the IEEE International Conference on
• For an image processing practitioner, Climb provides Image Processing (ICIP), pages 1941–1944, Hong Kong,
a set of built-in algorithms, composition tools such as Sept. 2010.
the chaining operator and morphers to simplify the [3] F. Meyer. Un algorithme optimal pour la ligne de
matching of processed data and algorithms IO. partage des eaux, pages 847–857. AFCET, 1991.
[4] N. Otsu. A threshold selection method from gray-level
• For an algorithm implementor, Climb provides a very histograms. IEEE Transactions on Systems, Man and
high-level domain model articulated around three lev- Cybernetics, 9(1):62–66, Jan. 1979.
els of abstraction. Genericity on values (RGB, Grayscale [5] J. Sauvola and M. Pietik¨inen. Adaptive document
a
etc.), genericity on structures (sites, site sets) and gener- image binarization. Pattern Recognition, 33(2):225–236,
icity on implementations thanks to properties. 2000.
[6] P. Soille. Morphological Image Analysis: Principles and
Applications. Springer-Verlag New York, Inc., Secaucus,
The level of abstraction provided by the library makes it
NJ, USA, 2 edition, 2003.
possible to implement algorithms without any prior knowl-
edge on the images to which they will be applied. Conversly
supporting a new image types should not have any impact
on the already implemented algorithms.
From our experience, it appears that Common Lisp is very
well suited to highly abstract designs. The level of generic-
ity attained in Climb is made considerably easier by Lisp’s
reflexivity and CLOS’ extensibility. Thanks to the macro
system, this level of abstraction can still be accessed in a rel-
atively simple way by providing domain-specific extensions
(e.g. the chaining operators).
Although the current implementation is considered to be
reasonably stable, Climb is still a very yound project and
should still be regarded as a prototype. In particular, noth-
ing has been done in terms of performance yet, as our main
focus was to design a very generic and expressive API. Fu-
ture work will focus on:
Genericity By providing new data types like graphs, and
enhancing the current property system.
Usability By extending the $ macro with support for au-
tomatic insertion of required morphers and offering a
GUI for visual programming of complex Image Pro-
cessing chains.](https://image.slidesharecdn.com/senta-12-els-120502060510-phpapp01/85/Generic-Image-Processing-With-Climb-4-320.jpg)
Generic Image Processing With Climb
- 1. Generic Image Processing With Climb Laurent Senta Christopher Chedeau Didier Verna senta@lrde.epita.fr christopher.chedeau@lrde.epita.fr didier.verna@lrde.epita.fr Epita Research and Development Laboratory 14-16 rue Voltaire, 94276 Le Kremlin-Bicêtre Paris, France ABSTRACT a dynamic language. Common Lisp provides the necessary Categories and Subject Descriptors flexibility and extensibility to let us consider several alter- I.4.0 [Image Processing And Computer Vision]: Gen- nate implementations of the model provided by Olena, from eral—image processing software; D.2.11 [Software Engi- the dynamic perspective. neering]: Software Architectures—Data abstraction, Domain- specific architectures First, we provide a survey of the algorithms readily available in Climb. Next, we present the tools available for compos- ing basic algorithms into more complex Image Processing General Terms chains. Finally we describe the generic building blocks used Design, Languages to create new algorithms, hereby extending the library. Keywords 2. FEATURES Generic Image Processing, Common Lisp Climb provides a basic image type with the necessary load- ing and saving operations, based on the lisp-magick library. 1. INTRODUCTION This allows for direct manipulation of many images formats, Climb is a generic image processing library written in Com- such as PNG and JPEG. The basic operations provided by mon Lisp. It comes with a set of built-in algorithms such as Climb are loading an image, applying algorithms to it and erosion, dilation and other mathematical morphology oper- saving it back. In this section, we provide a survey of the ators, thresholding and image segmentation. algorithms already built in the library. The idea behind genericity is to be able to write algorithms only once, independently from the data types to which they 2.1 Histograms will be applied. In the Image Processing domain, being fully A histogram is a 2D representation of image information. generic means being independent from the image formats, The horizontal axis represents tonal values, from 0 to 255 pixels types, storage schemes (arrays, graphs, dimensions) in common grayscale images. For every tonal value, the etc. Such level of genericity, however, may induce an impor- vertical axis counts the number of such pixels in the image. tant performance cost. Some Image Processing algorithms may be improved when the information provided by a histogram is known. In order to reconcile genericity and performance, the LRDE1 has developed an Image Processing platform called Olena2 . Olena is written in C++ and uses its templating system Histogram Equalization. An image (e.g. with a low con- abundantly. Olena is a ten years old project, well-proven trast) does not necessarily contain the full spectrum of in- both in terms of usability, performance and genericity. [2] tensity. In that case, its histogram will only occupy a nar- row band on the horizontal axis. Concretely, this means In this context, the goal of Climb is to use this legacy for that some level of detail may be hidden to the human eye. proposing another vision of the same domain, only based on Histogram equalization identifies how the pixel values are spread and transforms the image in order to use the full 1 EPITA Research and development laboratory range of gray, making details more apparent. 2 http://olena.lrde.epita.fr Plateau Equalization. Plateau equalization is used to re- veal the darkest details of an image, that would not oth- erwise be visible. It is a histogram equalization algorithm applied on a distribution of a clipped histogram. The pixels values greater than a given threshold are set to this thresh- old while the other ones are left untouched. 2.2 Threshold
- 2. Thresholding is a basic binarization algorithm that converts Mathematical Morphology can also be applied on grayscale a grayscale image to a binary one. This class of algorithms images, leading to news algorithms: works by comparing each value from the original image to a threshold. Values beneath the threshold are set to False in the resulting image, while the others are set to True. TopHat Sharpen details Laplacian operator Extract edges Climb provides 3 threshold algorithms: Hit-Or-Miss Detect specific patterns Basic threshold. Apply the algorithm using a user-defined 3. COMPOSITION TOOLS threshold on the whole image. Composing algorithms like the ones described in the previ- ous section is interesting for producing more complex image processing chains. Therefore, in addition to the built-in al- Otsu Threshold. Automatically find the best threshold value gorithms that Climb already provides, a composition infras- using the image histogram. The algorithm picks the value tructure is available. that minimizes the spreads between the True and False do- mains as described by Nobuyuki Otsu [4]. 3.1 Chaining Operators 3.1.1 The $ operator The most essential form of composition is the sequential one: Sauvola threshold. In images with a lot of variation, global algorithms are applied to an image, one after the other. thresholding is not accurate anymore, so adaptive threshold- ing must be used. The Sauvola thresholding algorithm [5] Programming a chain of algorithms by hand requires a lot picks the best threshold value for each pixel from the given of variables for intermediate storage of temporary images. image using its neighbors. This renders the code cumbersome to read an maintain. 2.3 Watershed To facilitate the writing of such a chain, we therefore pro- vide a chaining operator which automates the plugging of Watershed is another class of image segmentation algorithms. an algorithm’s output to the input of the next one. Such Used on a grayscale image seen as a topographic relief, they an operator would be written as follows in traditional Lisp extract the different basins within the image. Watershed al- syntax: gorithms produce a segmented image where each component is labeled with a different tag. ( image−s a v e ( algo2 Climb provides a implementation of Meyer’s flooding algo- ( algo1 rithm [3]. Components are initialized at the “lowest” region ( image−l o a d ‘ ‘ image . png ’ ’ ) of the image (the darkest ones). Then, each component is param1 ) extended until it collide with another one, following the im- param2 ) age’s topography. ‘ ‘ output . png ’ ’ ) 2.4 Mathematical Morphology Unfortunately, this syntax is not optimal. The algorithms Mathematical Morphology used heavily in Image Processing. must be read in reverse order, and arguments end up far It defines a set of operators that can be cumulated in order away from the function to which they are passed. to extract specific details from images. Inspired by Clojure’s Trush operator (->) 3 and the JQuery4 Mathematical Morphology algorithms are based on the ap- chaining process, Climb provides the $ macro which allows plication of a structuring element to every points of an im- to chain operations in a much clearer form. age, gathering information using basic operators (e.g. logical $ ( ( image−l o a d ‘ ‘ image . png ’ ’ ) operators and, or) [6]. ( a l g o 1 param1 ) ( a l g o 2 param2 ) ( image−s a v e ) ) Erosion and Dilation. Applied on a binary image, these algorithms respectively erodes and dilate the True areas in the picture. Combining these operators leads to new algo- 3.1.2 Flow Control rithms: In order to ease the writing of a processing chain, the $ macro is equipped with several other flow control operators. Opening Apply an erosion then a dilation. This removes the small details from the image and opens the True • The // operator splits the chain into several indepen- shapes. dent subchains that may be executed in parallel. 3 http://clojure.github.com/clojure/clojure. Closing Apply a dilation then an erosion. This closes the core-api.html#clojure.core/-> 4 True shapes. http://api.jquery.com/jQuery/
- 3. • The quote modifier (’) allows to execute a function Value morpher. A value morpher operates a dynamic trans- outside the actual processing chain. This is useful for lation from one value type to another. For instance, an RGB doing side effects (see the example below). image can be used as a grayscale one when seen through a morpher that returns the pixels intensity instead of their • The $1, $2, etc. variables store the intermediate results original color. It is therefore possible to apply a watershed from the previous executions, allowing to use them ex- algorithm on a colored image in a transparent fashion, with- plicitly as arguments in the subsequent calls in the out actually transforming the original image into a grayscale chain. one. • The # modifier may be used to disrupt the implicit chaining and let the programmer use the $1, etc. vari- ables explictly. Content morpher. A content morpher operates a dynamic translation from one image structure to another. For in- stance, a small image can be used as a bigger one when seen These operators allow for powerful chaining schemes, as il- through a morpher that returns a default value (e.g. black), lustrated below: when accessing coordinates outside the original image do- ( $ ( l o a d ‘ ‘ l e n a g r a y . png ’ ’ ) main. It is therefore possible to apply an algorithm built for images with a specific size (e.g. power of two), without (// having to extend the original image. ( ’ ( timer −s t a r t ) ( otsu ) Possible uses for content morphers include restriction (parts ’ ( timer −p r i n t ”Otsu ”) of the structures being ignored), addition (multiple struc- ( s a v e ” o t s u . png ”) ) ; $1 tures being combined) and reordering (structures order be- ing modified). ( ’ ( timer −s t a r t ) ( s a u v o l a ( box2d 1 ) ) 4. EXTENSIBILITY ’ ( timer −p r i n t ”Sauvol a 1 ”) The pixels of an image are usually represented aligned on ( s a v e ” s a u v o l a 1 . png ”) ) ; $2 a regular 2D grid. In this case, the term “pixel” can be interpreted in two ways, the position and the value, which ( ’ ( timer −s t a r t ) we need to clearly separate. Besides, digital images are not ( s a u v o l a ( box2d 5 ) ) necessarily represented on a 2D grid. Values can be placed ’ ( timer −p r i n t ”Sauvol a 5 ”) on hexagonal grids, 3D grids, graphs etc. In order to be ( s a v e ” s a u v o l a 5 . png ”) ) ) ; $3 sufficiently generic, we hereby define an image as a function from a position (a site) to a value: img(site) = value. (// (#( d i f f $1 $2 ) A truly generic algorithm should not depend on the under- ( save ‘ ‘ d i f f −otsu−s a u v o l a 1 . png ’ ’ ) ) lying structure of the image. Instead it should rely on higher (#( d i f f $1 $3 ) level concepts such as neighborhoods, iterators, etc. More ( save ‘ ‘ d i f f −otsu−s a u v o l a 5 . png ’ ’ ) ) precisely, we have identified 3 different levels of genericity, (#( d i f f $2 $3 ) as explained below. ( save ‘ ‘ d i f f −s a u v o l a 1 −s a u v o l a 5 . png ’ ’ ) ) ) ) 4.1 Genericity on values Genericity on values is based on the CLOS dispatch. Us- 3.2 Morphers ing multimethods, generic operators like comparison and As we saw previously, many Image Processing algorithms addition are built for each value type, such as RGB and result in images of a different type than the original image Grayscale. (for instance the threshold of a grayscale image is a boolean one). What’s more, several algorithms can only be applied on images of a specific type (for instance, watershed may 4.2 Genericity on structures only be applied to grayscale images). The notion of “site” provides an abstract description for any kind of position within an image. For instance, a site might Programming a chain of algorithms by hand requires a lot of be an n-uplet for an N-D image, a node within a graph, etc. explicit conversion from one type to another. This renders the code cumbersome to read an maintain. Climb also provides a way to represent the regions of an image through the site-set object. This object is used in or- To facilitate the writing of such a chain, we therefore provide der to browse a set of coordinates and becomes particularly an extension of the Morpher concept introduced in Olena handy when dealing with neighborhoods. For instance, the with SCOOP2 [1] generalized to any object with a defined set of nearest neighbors, for a given site, would be a list of communication protocol. Morphers are wrappers created coordinates around a point for an N-D image, or the con- around an object to modify how it is viewed from the outside nected nodes in a graph. world. 4.3 Genericity on implementations Two main categories of morphers are implemented: value Both of the kinds of genericity described previously allow morphers and content morphers. algorithm implementors to produce a single program that
- 4. works with any data type. Some algorithms, however, may Performance By improving the state of the current im- be implemented in different ways when additional informa- plementation, exploring property-based algorithm op- tion on the underlying image structure is available. For in- timization and using the compile-time facilities offered stance, iterators may be implemented more efficiently when by the language. we know image contents are stored in an array. We can also distinguish the cases where a 2D image is stored as a matrix or a 1D array of pixels etc. Climb is primarily meant to be an experimental plateform for exploring various generic paradigms from the dynamic Climb provides a property description and filtering system languages perspective. In the future however, we also hope that gives a finer control over the algorithm selection. This is to reach a level of usability that will trigger interest amongst done by assigning properties such as :dimension = :1D, :2D, Image Processing practitioners. etc. to images. When defining functions with the defalgo macro, these properties are handled by the dispatch system. 6. REFERENCES The appropriate version of a function is selected at runtime, [1] Th. G´raud and R. Levillain. Semantics-driven e depending on the properties implemented by the processed genericity: A sequel to the static C++ object-oriented image. programming paradigm (SCOOP 2). In Proceedings of the 6th International Workshop on Multiparadigm 5. CONCLUSION Programming with Object-Oriented Languages Climb is a generic Image Processing library written in Com- (MPOOL), Paphos, Cyprus, July 2008. mon Lisp. It is currently architected as two layers targeting [2] R. Levillain, Th. G´raud, and L. Najman. Why and e two different kinds of users. how to design a generic and efficient image processing framework: The case of the Milena library. In Proceedings of the IEEE International Conference on • For an image processing practitioner, Climb provides Image Processing (ICIP), pages 1941–1944, Hong Kong, a set of built-in algorithms, composition tools such as Sept. 2010. the chaining operator and morphers to simplify the [3] F. Meyer. Un algorithme optimal pour la ligne de matching of processed data and algorithms IO. partage des eaux, pages 847–857. AFCET, 1991. [4] N. Otsu. A threshold selection method from gray-level • For an algorithm implementor, Climb provides a very histograms. IEEE Transactions on Systems, Man and high-level domain model articulated around three lev- Cybernetics, 9(1):62–66, Jan. 1979. els of abstraction. Genericity on values (RGB, Grayscale [5] J. Sauvola and M. Pietik¨inen. Adaptive document a etc.), genericity on structures (sites, site sets) and gener- image binarization. Pattern Recognition, 33(2):225–236, icity on implementations thanks to properties. 2000. [6] P. Soille. Morphological Image Analysis: Principles and Applications. Springer-Verlag New York, Inc., Secaucus, The level of abstraction provided by the library makes it NJ, USA, 2 edition, 2003. possible to implement algorithms without any prior knowl- edge on the images to which they will be applied. Conversly supporting a new image types should not have any impact on the already implemented algorithms. From our experience, it appears that Common Lisp is very well suited to highly abstract designs. The level of generic- ity attained in Climb is made considerably easier by Lisp’s reflexivity and CLOS’ extensibility. Thanks to the macro system, this level of abstraction can still be accessed in a rel- atively simple way by providing domain-specific extensions (e.g. the chaining operators). Although the current implementation is considered to be reasonably stable, Climb is still a very yound project and should still be regarded as a prototype. In particular, noth- ing has been done in terms of performance yet, as our main focus was to design a very generic and expressive API. Fu- ture work will focus on: Genericity By providing new data types like graphs, and enhancing the current property system. Usability By extending the $ macro with support for au- tomatic insertion of required morphers and offering a GUI for visual programming of complex Image Pro- cessing chains.