GA RELAXATION LABELING

Abstract

This paper presents a study of relaxation labeling of line images using Genetic Algorithms (GA). The proposed technique considers the labeling problem as two combination optimization problems and uses a GA search to obtain an optimal solution. A novel multi-GA system approach is proposed in the first optimization stage to produce a GA population for optimization by the second stage. Image partitioning for large images prior to labeling followed by reconstitution after labeling with appropriate edge processing was also implemented to achieve better optimization and shorter processing time. The proposed scheme was tested on several noiseless line drawings and achieved an average labeling success rate of over 80% for the test images used. Enlarging the same images by a factor of three while keeping the line thickness constant, led to average labeling success rates of almost 95%. Image restoration from random noise degradation was also achieved to some extent. The noise rejection capability of the scheme proved to be quite good; between 36.5% and 96.9% for different line images. The proposed labeling scheme was found to perform comparatively well compared with a popular deterministic labeling method, being superior in some cases but with the disadvantage of significantly longer processing time.