Fused shape features based on gradient and local phase in color domain

Abstract

In this paper we present a new descriptor based on shape features for human detection. The shape features are extracted based on both, the image gradients, and the local phase in color space. The fusing of these complementary information yields to capture a broad range of the human appearance details that improves the detection accuracy. The proposed features are formed by computing the phase congruency of the three-color channels in addition to the gradient magnitude and orientation for each pixel in the image with respect to its neighborhood. Only the maximum phase congruency values are selected from the corresponding color channels. The histogram of oriented phase and the histogram of oriented gradients for the local regions of the image, are determined. These histograms are concatenated to construct the proposed descriptor and it is named as Fused Gradients and local Phase in Color space (FGPC). Several experiments were performed to test and evaluate the detection performance of the proposed descriptor. A linear support vector machine (SVM) classifier is used to train the pedestrians. The experimental results show that the human detection system based on the proposed features has less error rates and better detection performance over a set of state of the art feature extraction methodologies.