CS292 Computational Vision and Language
Segmentation and Region Detection
Introduction
• All pixels belong to a region, which can be– an object– part of object– background
• Find region– By finding constituent pixels in a region– By finding boundary pixels
Image Segmentation Task
• To divide the image into regions or segments, each of which is in some sense homogeneous, but the union of adjacent segments is not homogeneous in the same sense.
• Homogeneity here is characterized by some properties like– smoothly varying intensity, similar
statistics, or colour.
Region Detection
• A set of pixels P
• An homogeneity predicate H(P)
• Partition P into regions {R}, such that
iR
i1
n
P
H iR True
H iR jR False
Image Segmentation• Many techniques including
– Non-contextual technique: thresholding– Contextual techniques:
• region-based
• connectivity-preserving relaxation methods.
– Other methods: Image primitive based
• Mixture of all these
Threshold techniques
• make decisions based on local pixel information– are effective when the intensity levels of the objects fall
squarely outside the range of levels in the background.
Global thresholds
• Compute threshold from whole image
• Incorrect in some regions
Local thresholds
• Divide image into regions
• Compute threshold per region
• Merge thresholds across region boundaries
Contextual techniques
• Contextual techniques take into account the fact that pixels belonging to a single object are close to one another.
• Approaches to contextual segmentation are based on the concept of discontinuity or concept of similarity. – detecting abrupt changes- edge detection techniques,
– or to create uniform regions directly,
• Discontinuity and similarity approaches mirror one another, in the sense that completion of boundary is equivalent to breaking one region into two.
Region Growing
• All pixels belong to a region
• Select a pixel
• Grow the surrounding region
(we will practise this in lab class)
Slow Algorithm
• If a pixel is– not assigned to a region– adjacent to region– has colour properties not different to region’s
• Then– Add to region– Update region properties
Split and Merge
• Initialise image as a region
• While region is not homogeneous– split into quadrants and examine homogeneity
Recursive SplittingSplit(P){ If (!H(P)) {
P subregions 1 … 4;Split (subregion 1);Split (subregion 2);Split (subregion 3);Split (subregion 4);
}}
Recursive Merging
• If adjacent regions are– weakly split
• weak edge, depending on defined criteria
– similar• similar greyscale/colour properties
• Merge them
Edge Following
• Detection– finds candidate edge pixels
• Following– links candidates to form boundaries
Representing Regions
• Constituent pixels
• Boundary pixels
Based on both regions and edges
Based on the combination of colour and texture
Active Contour Model- Snake
• A connectivity-preserving relaxation-based segmentation method, - active contour model – snake– The main idea is to start with some initial boundary shape
represented in the form of spline curves, and iteratively modify it by applying various shrink/expansion operations according to some energy function.
• Concepts involved– Image gradient– Smooth operation– Histogram equalization– Energy functions
Snakes, Active/Dynamic Contours
• Borders follow outline of object
• Outline obscured?
• Snake provides a solution
Algorithm
• Snake computes smooth, continuous border
• Minimises– length of border– curvature of border
• Against an image property– gradient?
EEEE imagecurvaturelengthtot
Minimisation
• Initialise snake
• Integrate energy along it
• Iteratively move snake to global energy minimum
Active Contour Method
Case study next week, notes will be given during the lecture
Summary
• Image segmentation
• Region detection– growing– edge following
