Region Based Colour Image Segmentation Using Bio Inspired Optimization Algorithms

Abstract

ABSTRACT newlineImage processing is a very tedious and fundamental task in many machine vision applications. Segmentation of images helps one to identify objects present in the image. Identification of objects is made easier by isolating (segmenting) the region of interest. A number of researches have been made on image segmentation. In all algorithms, the necessary and sufficient condition taken under consideration is to divide the image into several homogeneous regions. This is the most important step which is used to improve the overall performance of the algorithm. newlineTwo bio-inspired optimization algorithms are proposed to segment color images. Initially, a segmentation algorithm based on region growing technique is proposed which is then optimized with Cuckoo Search (CS) and FireFly (FF) algorithms. The performance of the segmented outputs after optimization is satisfactory. newlineIn the proposed Dynamic Modified Region Growing algorithm, regions are grown from the initial seeds. The quality of final segmentation depends on the location of the initial seed points. This algorithm initially searches for regions without edges, and these pixels are selected as the initial seed pixels. After selecting the initial seeds regions are grown by appending neighboring pixels with the initial seed, pixel based on intensity and orientation. This forms the initially segmented region grown image. The presence of regions with patterns that contain textures results in oversegmentation, and the interpretation of the edges is also difficult. Hence texture characterization is also done to improve the performance in the case of textured images. Texture provides random information, and this randomness is evaluated using entropy measurement. Finally, region merging is done to avoid oversegmentation problem. Oversegmentation may be due to the presence of texture or due to the small variation in color. Hence merging of regions is done by considering both the color and the texture information. newlineQuality of the segmentation can be further improved by m