Abstract:
COLOR IMAGE SEGMENTATION: A CROSS POINT
APPROACH.
Segmentation is one of the most important steps in image processing. In the hierarchy
of image engineering it lies at the center which binds the low level image processing
techniques based on pixel to the higher level of image understanding layer and object
recognition. Many techniques are developed that segment images based on various
mathematical features of images. My main focus in this work was to develop an
efficient algorithm to segment color images.
In this dissertation a different technique is presented where human perception is the
main motivation. Here a combination of histograms pairs is used to extract segment
points. Initially color image, stored in any format, converted in RGB. Then this RGB
image is converted in inverse RGB format which is used to extract histogram curves.
After smoothing these histogram curves, cross points a taken out from different
combinations of the cures that are ranked accordingly. These cross-points are
optimized by applying filters and some variables to control the quality of the
segments.
During the process of segmentation not only the value of pixels are considered but the
neighbors of the pixels and their association are also used which give uniqueness to
this algorithm.
Image set provided by the Berkeley University of California is used to train and test
the proposed algorithm. For the validation of the results human judgments are also
used along with the statistical tools like geometric mean, variance, and linear
correlation. 100 test images are used for this purpose.
