Presentation is loading. Please wait.

Presentation is loading. Please wait.

Machine Vision ENT 273 Lecture 4 Hema C.R.

Similar presentations


Presentation on theme: "Machine Vision ENT 273 Lecture 4 Hema C.R."— Presentation transcript:

1 Machine Vision ENT 273 Lecture 4 Hema C.R.
Regions and Segmentation in Images Lecture 4

2 Road Map Regions Region Segmentation Thresholding Methods
Region Representation Region Merging Region Splitting Region Growing Hema ENT 273 Lecture 4

3 Introduction First step in image analysis is to partition the image into object or region. There are 2 ways of partitioning an image Region-based segmentation Boundary estimation using edge detection Hema ENT 273 Lecture 4

4 Regions A region is a group of connected pixels with similar properties Regions may correspond to objects in a scene and hence are important in image interpretation Hema ENT 273 Lecture 4

5 Segmentation Segmentation A method to partition an image into sub images is called regions Each region is an object candidate Segmentation is a very important step in understanding images Principal of segmentation Value similarity Spatial proximity Hema ENT 273 Lecture 4

6 Region Segmentation Given a set of image pixels I and a homogenous predicate P(.), a partition of the image I into a set of n region Ri The predicate properties are any region satisfy the predicate and any two adjacent regions cannot merged Hema ENT 273 Lecture 4

7 Region Segmentation Thresholding is the simplest region segmentation technique Simple segmentation is conversion of a grayscale image to a binary image where image is partitioned into two sets. Hema ENT 273 Lecture 4

8 Automatic Thresholding
A thresholding scheme that uses knowledge of objects such as Intensity levels of objects Size of objects Fraction of image occupied by objects Types of different objects in image to select a proper threshold value for each image without human intervention is called an automatic thresholding scheme Hema ENT 273 Lecture 4

9 Common Approaches to Automatic Thresholding
P-tile method Mode method Iterative Method Adaptive Method Variable Method Double Thresholding Method Hema ENT 273 Lecture 4

10 Common Approaches to Automatic Thresholding
P- Tile Method Use size or area of desired object to threshold an image, threshold chosen to assign p percent of pixels to the object. Hema ENT 273 Lecture 4

11 Common Approaches to Automatic Thresholding
Mode Method When objects and background have two different gray levels , then threshold can be determined from the valley points of the histogram Hema ENT 273 Lecture 4

12 If there are n objects with various peak
Hema ENT 273 Lecture 4

13 Common Approaches to Automatic Thresholding
Iterative Threshold Selection Starts with approximate threshold and successively refined based on results Hema ENT 273 Lecture 4

14 Common Approaches to Automatic Thresholding
Adaptive Thresholding Used in cases of uneven illumination, analyses sub images to obtain threshold for sub image Hema ENT 273 Lecture 4

15 To overcome uneven distribution of background, we divide the image to sub-images
Hema ENT 273 Lecture 4

16 Common Approaches to Automatic Thresholding
Variable Thresholding Used in uneven illumination, approximates intensity values of image by a simple planar function to find threshold of the image. Hema ENT 273 Lecture 4

17 Common Approaches to Automatic Thresholding
Double Thresholding Uses two thresholds to segment an image, first T to segment core image and second T to select pixels connected to core. Hema ENT 273 Lecture 4

18 Region Representation
Three Types Array Representation Hierarchical Representation Symbolic Representation Hema ENT 273 Lecture 4

19 Array Representation An array of the same size of the original image is used to indicate the region to which a pixel belongs Membership arrays called masks are used to indicate which pixels belong to that region E.g.. Background pixels in a binary image Hema ENT 273 Lecture 4

20 Hierarchical Representation
Hierarchical representation make it possible to use algorithms which decide a strategy for processing on the basis of relatively small quantities of data. Allows representation of images in multiple resolutions Pyramid An image having one degree smaller resolution in a pyramid contains four times less data, so that it is processed approximately four times as quickly Pyramid Hema ENT 273 Lecture 4

21 Hierarchical Representation
Pyramids-nxn image is reduced to k versions of a power of 2, n/2xn/2, n/4xn/4x.....1x1 Hema ENT 273 Lecture 4

22 Hierarchical Representation
Quad Tree Quad tree is an extension of pyramids for binary images Contains three nodes White Black Gray Obtained by recursive splitting of an image When all points in sub region are either black or white then region is no longer considered for splitting Quad Tree Hema ENT 273 Lecture 4

23 r Hema ENT 273 Lecture 4

24 Data Structure for Segmentation
Data structure is to allow easy manipulation of region characteristic One of the technique is through Picture Trees which emphasize the inclusion of a region within another region Hema ENT 273 Lecture 4

25 Region Merging Combines regions considered to be similar
Important - to determine similarity between regions Region based merging Two approaches Compare mean intensities with some threshold Probability distribution of intensities Removing weak edges Common boundary between two regions is dissolved if boundary is weak and gray value of resulting boundary is not much effected. A weak boundary is one for which the intensities on either side differ by less than a threshold T Hema ENT 273 Lecture 4

26 Region Splitting If some property of a region is not constant then the region should be split Important : where to split? Segmentation can be refined by combining split and merge operations Hema ENT 273 Lecture 4

27 Region Splitting (Cont.)
The basic idea of region splitting is to break the image into a set of disjoint regions which are coherent within themselves: Initially take the image as a whole to be the area of interest. Look at the area of interest and decide if all pixels contained in the region satisfy some similarity constraint. This process continues until no further splitting occurs. In the worst case this happens when the areas are just one pixel in size. This is a divide and conquer or top down method. Hema ENT 273 Lecture 4

28 Split and Merge Intensity based segmentation results in too many regions These will result in false boundaries and spurious regions Subdivide an image initially into a set of arbitrary, disjoint regions and then merge and/or split the regions in an attempt to satisfy the necessary conditions Images are to be refined or reformed include: Merge adjacent regions with similar characteristic Remove questionable edges Use topological properties of the regions Use shape information of objects Use semantic information about the scene Hema ENT 273 Lecture 4

29 Let R represent entire image region and select a predicate P
(1) Split into four disjoint quadrants any region Ri for which P(Ri) = FALSE (2) Merge any adjacent regions Rj and Rk for which P(Rj U Rk) = TRUE (3) Stop when no further merging or splitting is possible Hema ENT 273 Lecture 4

30 Hema ENT 273 Lecture 4

31 Example: Split and merge
Hema ENT 273 Lecture 4

32 Region Growing Selects a seed region in an image
Merges adjacent similar regions to the seed Repeats till all regions except regions smaller than a given criteria Hema ENT 273 Lecture 4

33 Region Growing An initial set of small areas are iteratively merged according to similarity constraints. Start by choosing an arbitrary seed pixel and compare it with neighbouring pixels Region is grown from the seed pixel by adding in neighbouring pixels that are similar, increasing the size of the region. When the growth of one region stops we simply choose another seed pixel which does not yet belong to any region and start again. This whole process is continued until all pixels belong to some region. A bottom up method. Hema ENT 273 Lecture 4

34 Example of region growing
Hema ENT 273 Lecture 4

35 Example: Application of region growing to weld inspection
Hema ENT 273 Lecture 4

36 Machine Systems ENT 273 End of Lecture 4


Download ppt "Machine Vision ENT 273 Lecture 4 Hema C.R."

Similar presentations


Ads by Google