Robust Face Detection Based on Enhanced Local Sensitive Support Vector Machine

Abstract

In recent years, local classifiers have obtained great success in classification task due to its powerful discriminating ability on local regions. Based on it, we employ a locality-sensitive SVM (LSSVM) to build a local model on each local region to solve the problem of large intra-class variances between different face images. On the other hand, the use of SVM with local kernels was presented. Compared with the conventional global kernel, it’s more robust since it can utilize the local features which are influenced only specific parts under partial occlusion. So in order to detect face effectively, we want to utilize the global and local features of face comprehensively. Thus we combine the global and local kernels and apply the combination kernel to the LSSVM algorithm, proposing a robust face detection algorithm. Extensive experiments on the widely used CMU+MIT dataset and FDDB dataset demonstrate the robustness and validity of our algorithm.