Image segmentation
Download as PPTX, PDF2 likes4,703 views
The document discusses various methods of image segmentation, which is the process of partitioning an image into homogenous regions. It covers edge-based and region-based segmentation techniques such as region growing and split-and-merge, as well as active contour models (snakes) and their applications in medical imaging. The conclusion emphasizes the complexity of image analysis in medical contexts and the necessity of incorporating prior knowledge to improve segmentation outcomes.
Image segmentation
- 2. 2 Outline Introduction Edge-based segmentation Region-based segmentation – Region Growing – Split-and-merge Active contour models (snakes) Application in medical imaging Conclusion Assignment 2
- 3. 3 Introduction What is Image segmentation ? – The different partitioning of an image into non- overlapping, constituent regions which are homogeneous with respect to some characteristic such as intensity or texture. – Each of such homogeneous regions may represent an object.
- 4. 4 Introduction The shape of an object can be described in terms of: Its boundary – requires image edge detection The region it occupies – requires image segmentation in homogeneous regions, Image regions generally have homogeneous characteristics (e.g. intensity, texture) Segmentation methods are then classified into – Edge-based – Region-based
- 5. 5 Edge-based Segmentation Emphasis: – Determine the boundaries that separate regions Common approaches – Find edge points in the image Gradient based methods Second order methods – Linking these points in some way to produce description of edges in terms of lines, curves etc
- 6. 6 Detecting Edge points-Gradient based methods An edge point is a point where a discontinuity in gradient occurs across some line Different types of discontinuity are shown in the figure
- 7. 7 Gradient based methods (cont.) The gradient is a vector whose components measure how rapidly pixel values are changing with distance, in the x and y directions
- 8. 8 Gradient based methods (cont.) Considering dx=dy=1 (pixel spacing) and considering a point (i,j) Δx = f(i+1,j) – f(i,j) Δy = f(i,j+1) – f(i,j) Δx and Δy can be calculated by convolving the image with convolution masks
- 9. 9 Gradient based methods (cont.) To approximate the gradient along directions at 45o and 135o to the axes respectively, known as Roberts edge operator. The corresponding convolution masks are Other 3x3 edge operators can be used such as Sobel and Canny
- 10. 10 Example Original image Image produced by the horizontal gradient calculation Image produced by the vertical gradient calculation
- 11. 11 Example Gradient image formed by combining horizontal and vertical gradient detection Gradient image is produced using the magnitude M form M = | Δx | + | Δy | The gradient direction θ is equal to Θ = tan-1 (Δy / Δx )
- 12. 12 Implementation using Matlab Read image and display it. I = imread('coins.jpg'); imshow(I) Apply the Sobel and Canny edge detectors to the image and display them BW1 = edge(I,'sobel'); BW2 = edge(I,'canny'); imshow(BW1) figure, imshow(BW2) Original image “coins.jpg”
- 13. 13 Second order methods Laplacian operator The laplacian function is defined by Laplacian mask
- 14. 14 Edge Linking Edge detectors yield pixels that lie on edges The objective is to replace many points on edges with real edges. Edge linking can be performed by: – Local edge linkers – where edge points are grouped according to their relationships with the neighboring edge points. – Global Edge Linkers – Hough transform
- 15. 15 Hough Transform Allows recognition of global patterns in an image Finds curves like straight lines, circles, etc Suppose that we are looking for straight lines in an image – If we take a point (x',y') in the image, all lines which pass through that pixel have the form y’ = mx’ +c
- 16. 16 Hough Transform This equation can be written as c = -x’m + y’ where x’,y’ are constants and m,c varies Each different line through the point (x',y') corresponds to one of the points on the line in (m,c) space Lines through a point
- 17. 17 Hough Transform All pixels which lie on the same line in (x,y) space are represented by lines which all pass through a single point in (m,c) space. The single point through which they all pass gives the values of m and c in the equation of the line y=mx+c.
- 18. 18 Hough Transform The y=mx+c form for representing a straight line breaks down for vertical lines, when m becomes infinite. To avoid this problem, it is better to describe straight lines in the form of x cos θ + y sin θ = r i.e. a point in (x,y) space is now represented by a curve in (r,θ) space rather than a straight line
- 19. 19 Hough Transform To detect straight lines in an image 1. Quantize (m,c) space into a two-dimensional array A for appropriate steps of m and c. 2. Initialize all elements of A(m,c) to zero. 3. For each pixel (x',y') which lies on some edge in the image, add 1 to all elements of A(m,c) whose indices m and c satisfy y'=mx'+c. 4. Search for elements of A(m,c) which have large values -- Each one found corresponds to a line in the original image.
- 20. 20 Hough Transform To find circles, with equation (x – a)2 + (y – b)2 = r2 – Every point in (x,y) space corresponds to a surface in (a,b,r) space (as we can vary any two of a, b and r, but the third is determined by the equation of the circle). – The basic method is, thus, modified to use a three-dimensional array A(a,b,r), – All points in it which satisfy the equation for a circle are incremented. The technique takes rapidly increasing amounts of time for more complicated curves as the number of variables (and hence the number of dimensions of A) increases
- 21. 21 Outline Introduction Edge-based segmentation Region-based segmentation – Region Growing – Split-and-merge Active contour models (snakes)
- 22. 22 Region Growing A simple approach to image segmentation is to start from some pixels (seeds) representing distinct image regions and to grow them, until they cover the entire image Before assigning a pixel x to a region Ri(k), check if the region is homogeneous: i.e. H(Ri(k) U {x}) = TRUE
- 23. 23 Region Growing The arithmetic mean M and standard deviation sd can be used to decide if merging two regions R1,R2 is allowed if |M1 – M2| < (k)*sd(i) , i = 1, 2 , merge the two regions where k is a certain threshold
- 24. 24 Split-and-Merge The opposite approach to region growing is region splitting. The approach starts with the assumption that the entire image is homogeneous If the entire image is not homogeneous, the image is split into four sub images This splitting procedure is repeated recursively until the image is split into homogeneous regions
- 25. 25 Split-and-Merge Since the procedure is recursive, it produces an image representation that can be described by a tree whose nodes have four children each Such a tree is called a Quadtree.
- 27. 27 Example Image and a representation of its quadtree decomposition
- 28. 28 Split-and-Merge Splitting techniques create regions that may be adjacent and homogeneous, but not merged. Split and Merge method is an iterative algorithm that includes both splitting and merging at each iteration. It produces more compact regions than the splitting algorithms
- 29. 29 Split-and-Merge Algorithm If a region R is inhomogeneous (H(R)= False) then split into four sub regions If two adjacent regions Ri,Rj are homogeneous (H(Ri U Rj) = TRUE), merge them Stop when no further splitting or merging is possible
- 30. 30 Outline Introduction Edge-based segmentation Region-based segmentation – Region Growing – Split-and-merge Active contour models (snakes) Application in medical imaging Conclusion
- 31. 31 Active Contour Models (Snakes) First introduced in 1987 by Kass et al, and gained popularity since then. Represents an object boundary as a parametric curve. An energy function E is associated with the curve. The problem of finding object boundary is an energy minimization problem.
- 32. 32 Framework for snakes A higher level process or a user initializes any curve close to the object boundary. The snake then starts deforming and moving towards the desired object boundary. In the end it completely “wraps” around the object. (Digram courtesy “Snakes, shapes, gradient vector flow”, Xu, Prince)
- 33. 33 Snakes Contour possesses an energy (Esnake) which is defined as the sum of the three energy terms. where Einternal represents the internal energy of the spline due to bending, Eexternal denotes image forces, and Econstraint denotes external constraint forces. The energy terms are defined such that the final position of the contour will have a minimum energy (Emin) Therefore the problem of detecting objects reduces to an energy minimization problem. int intsnake ernal external constraE E E E
- 34. 34 Outline Introduction Edge-based segmentation Region-based segmentation – Region Growing – Split-and-merge Active contour models (snakes) Application in medical imaging Conclusion
- 35. 35 Application in medical imaging In-vivo segmentation: automating or facilitating the delineation of anatomical structures and other regions of interest Segmentation methods vary widely depending on the specific application and imaging modality. There is currently no single segmentation method that yields acceptable results for every medical image. Selecting an appropriate approach to a segmentation problem can be a difficult dilemma.
- 36. 36 Example In reconstructing a 3D model of a prostate, the capsule contour needs to be extracted from the slices’ images Capsule contour
- 37. 37 Example The capsule consists of collagen fibers tissues that appear under the microscope as wavy lines (see figure) However, the capsule line is sometimes unrecognizable because of the naturally occurring intrusion of muscle into the prostate gland which makes the segmentation problem more challenging.
- 38. 38 Summary Introduction Edge-based segmentation Region-based segmentation – Region Growing – Split-and-merge Active contour models (snakes) Application in medical imaging Conclusion
- 39. 39 Conclusion Edge detection and region growing algorithms are very popular in most commercial image analysis tools The reason is that they are simple to understand and to implement, and they are very generic, as they do not assume specific knowledge about the objects to be analyzed. These methods are often the starting point for more sophisticated model-based methods
- 40. 40 Conclusion For complex image data, such as medical images, their usefulness is quite limited. Effective image analysis methods must incorporate a priori knowledge of the considered structures such as photometric properties (object intensity, contrast, texture); geometric properties (position, shape, motion, deformation); and the context, such as the relative position with respect to the other objects in the neighborhood