2. THE DIFFERENCE IMAGE:
There are two major approaches to extracting two-dimensional motion from image sequential optical flow and motion correspondence. Simple subtraction of images acquired at different instants in time makes motion detection possible, when there is a stationary camera and constant illumination. Both of these conditions are satisfied in the areas of application of our system.
A difference image is nothing but a binary image d (i , j) where non-zero values represent image areas with motion, that is areas where these was a substantial difference between gray levels in consecutive images p1 and p2:
d (i, j) = 0 if  p1 (i, j) – p2 (i, j)  <= ε
= 1 otherwise
Where ε is a small positive number. The figure 3 shows the resultant image obtained by subtracting the images 1 and 2. The threshold level used in the system is 0.8, which is found to be sufficient for obtaining a good binary difference image

2.1.1. Complete and partial segmentation

Complete segmentation results in a set of disjoint region corresponding uniquely with objects in the input image. In partial segmentation, the regions may not correspond directly with the image objects.
If partial segmentation is the goal, an image is divided into separate regions that are homogeneous with respect to a chosen property such as brightness, color, reflectivity, texture etc.
Segmentation methods can be divided into three groups according to the dominant features they employ. First is global knowledge about an image or its past; edge-based segmentation forms the second group and region based segmentation, the third. In the second and third group each region can be represented by its closed boundary and each closed boundary describes the region. Edge based segmentation methods find the borders between regions while region based methods construct regions directly.
Region growing techniques are generally better in noisy images where borders are not very easy to detect. Homogeneity is an important property of regions and is used as the main segmentation criterion in region growing, where the basic idea is to divide an image into to zones of maximum homogeneity.
A complete segmentation of an image R is a finite set of regions R1...Rs ,
s
R = U Ri Ri ∩ Rj = Φ i ≠ j
i =1

Further, for region-based segmentation, the conditions need to be satisfied.

H (Ri) = TRUE i = 1,2,s
H (Ri U Rj) = FALSE i≠j

Ri adjacent to Rj where S is the total number of regions in an image and H(Ri) is a binary homogeneity evaluation of region Ri. Resulting regions of the segmented image must be both homogeneous and maximal where ‘maximal’ means that the homogeneity criterion would not be true after merging a region with any adjacent region.

2.2. Region merging and splitting
The basic approaches to region-based segmentation are
• Region Merging
• Region Splitting
• Split-and-Merge processing.

Region merging starts with an over segmented image and merges similar or homogeneous regions to form larger regions until no further merging are possible. Region splitting is the opposite of region merging. It begins with an under segmented image where the regions are not homogeneous. The existing image regions are sequentially split to form regions properly.



2.3. Region growing and segmentation

Our system uses the region growing segmentation method to video the image in to regions. In region growing segmentation, a seed point is first chosen in the image. Then the eight neighbours of the pixel are checked for a specific threshold condition. If the condition is satisfied it is incorporated as part of the region. This process is repeated for each of the eight neighbours and this continues until every pixel has been checked, and the whole image has been segmented into regions.
In our system, the MATLAB function ‘bwlabel’ which performs region-growing segmentation. This function accepts the image to be segmented as input and returns a matrix representing the segmented image along with the number of segments. It is to be noted that the image at this stage of processing is a binary image with only two levels-black (1) and white (0).

2.3.1. Segmentation algorithm

 AN INITIAL SET OF SMALL AREAS ARBITERATIVELY MERGED ACCORDING TO SIMILARITY CONSTRAINTS.
 START BY CHOOSING AN ARBITRARY SEED PIXEL, COMPARE IT WITH NEIGHBOURING PIXELS.
 REGION IS GROWN FROM THE SEED PIXEL BY ADDING IN NEIGHBOURING PIXELS THAT ARE SIMILAR, INCREASING THE SIZE OF 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.
 THE WHOLE PROCESS IS CONTINUED UNTIL ALL PIXELS BELONG TO SOME REGION.