Classifier based on GA-optimized choquet integrals and its application on foreground detection 1

Abstract

Abstract

A novel model based on nonlinear integrals is developed for the foreground and background detection. The nonlinear integral based on fuzzy measures, or its generalization, efficiency measure, is modeled as an aggregation tool to fuse the texture and color features of pixels. By setting suitable threshold value, the fusing result is represented as a two-class classifier to determine whether the pixels being considered belong to foreground or background. An optimization program based on genetic algorithm is proposed to retrieve the critical parameters of the efficiency measure with respect to which the nonlinear integral is defined and the threshold value to classify foreground and background. This method can handle various small variations of background objects and support sensitive detection of moving targets. Experiments results indicate that foreground and background can be separated correctly by using this new model and relevant algorithm. Comparisons with some existing models also verify the performance of the model being presented.