Fuzzy fusion for skin detection

Complex image processing tasks rarely succeed through the application of just one methodology. The implementation of different methodologies, whose treatment of the input images is complementary, can help in the successful attainment of the system goal. The result of the complementary procedures has to be eventually fused in order for the system to improve the result of each methodology taken on its own. Computer vision systems mostly employ simple fusion strategies for this aim. This simplicity downplays the relevance of the fusion stage. The paper presents a framework for skin detection, a pre-processing task very useful in application fields like video surveillance, human-machine interface, and cyber-crime prosecution. The framework is based on the employment of the fuzzy integral, which subsumes the performance of more simple fusion operators. As it is shown herein the framework manages therefore to cope with the complexity of skin detection under changing illumination conditions. The performance evaluation of the framework is undertaken on hand of a benchmark database in the paper.