A new approach for similar images using game theory

A Content Based Image Retrieval (CBIR) system is the result of the combination of computer vision techniques such as segmentation, extraction and classification in order to detect similar images. In our work we propose a CBIR system based on global descriptors, multiple clustering and on multi-criteria optimization problems using game theory. We propose to determine the similar images as a Nash Equilibrium. We tackle the problem by splitting the optimization variables into three players. The first player minimizes its objective function using the first strategy (Color description), the second player by using the second strategy (Gist descriptor) and the third one by using the third strategy (Zernike descriptor). After using the K-means algorithm to divide each level of description on K clusters, we use the Nash equilibrium to detect the classes of membership of the searched image.