Scene Image Classification Based On D-S Evidence Theory And SVM

How to get computer recognize different scenes images in human's manner, which is the content of scene images classification research. Including more than five kinds of perceptional characterization, Gist describes the global characterization well Based on the magnitude of local gradient and the image edge information, PHOG describe part-size characterization better. If the information of Gist and PHOG could be combined to make the decision for classification, the effect for classification would improve obviously. As an imprecise reasoning theory in dealing with multi-source information, D-S evidence theory could not only deal with the uncertainty due to the inaccuracy by knowledge, but also be able to deal with uncertainty because of unawareness. In this paper, two classification method fusing Gist and PHOG of scene images is proposed based on D-S evidence theory.For the first method, a new characterization combined Gist and PHOG is proposed. Get the Gist and PHOG of scene images, fuse both of them based on D-S evidence theory, and then input the fused characterization into the SVM to classify. Two experiments are conducted to prove the efficiency of this method.For the second method, Gist vector and PHOG vector of scene images test set have been conducted experiments with SVM respectively, data of result for experiments are used to make the decision for classification based on the D-S evidence theory. Experiment is conducted to prove the efficiency as well.Besides, the choice of different SVM kernel function and the setting for SVM kernel parameter have been studied. The genetic algorithm has been used to study the optimization parameter choosing. The improved genetic algorithm is further proposed.