Multi-objective Estimation Of Distribution Algorithms For Hand Region Segmentation

Hand region segmentation is segmenting a hand region from an image.It is essentially a multi-objective optimization problem.Evolutionary optimization method is a probabilistic search method based on population,and is inspired by natural biological evolution.It can efficiently solve many actual optimization problems.Among many evolutionary optimization methods,estimation of distribution algorithm estimates the probability distribution of a dominant population,obtains a new population by sampling this probability distribution in search space,and is an efficient evolutionary optimization method.Although estimation of distribution algorithm is successfully used to solve many actual optimization problems,research work relating to multi-objective estimation of distribution algorithms for hand region segmentation is very little.Thus,by sufficiently using the domain knowledge of the hand region segmentation problem,this paper studies the estimation of distribution algorithm for efficiently solving this problem.In order to use estimation of distribution algorithm to solve the hand region segmentation problem,a multi-objective optimization model of the hand region segmentation problem is firstly presented.We hope that by solving this optimization model,the pixels whose colors are closest to skin colors are selected from an image,and a hand region is formed.According to the obtained hand region,the position of each fingertip in the hand region can be further obtained.In order to use estimation of distribution algorithm to solve the fingertip localization problem,a multi-objective optimization model of fingertip localization is presented.We hope that by solving this optimization model,the pixels which have some features are selected,and the fingertips are formed.When using estimation of distribution algorithm to solve the optimization model of hand region segmentation,according to that the hand pixels distribute around several segments,a segment probability distribution model of candidate solutions is firstly built,and candidate solutions are obtained by sampling around the segments.Secondly,in order to increase the accuracy of hand region segmentation,according to that each coordinate of a hand pixel distributes in an interval,an interval probability distribution model of candidate solutions is built,and candidate solutions are sampled in several intervals.Then,an interval probability distribution model and a segment probability distribution model are built in the first phase and the second phaserespectively,candidate solutions are sampled based on different models,and a two-phase estimation of distribution algorithm for solving the above optimization problem is given.Finally,the segment model is replaced by a Guass model,candidate solutions are obtained by Guass sampling,and an improved two-phase estimation of distribution algorithm is given.The proposed method is used to solve many hand region segmentation problems,and is compared by the existing methods.The experimental results demonstrate that the proposed method can efficiently segment the hand region.Based on the obtained hand region,when using estimation of distribution algorithm to solve the fingertip localization problem,according to that the fingertip pixels distribute around several centers,a multi-point probability distribution model of candidate solutions is built,and candidate solutions are obtained by sampling around the points.Many experimental results demonstrate that the proposed estimation of distribution algorithm can efficiently localize the fingertipsThe work in this paper provides an efficient method of solving the hand region segmentation problem,enriches the theoretical basis of estimation of distribution algorithm,and expands the application scope of the method.Furthermore,it provides a valuable reference for solving other practical complex optimization problems.Therefore,it has an important theoretical significance and a practical value.