Contourlet-Based Fusion Method for Video Surveillance Using the Pulse Coupled Neural Networks Model

Abstract

To improve the quality of images in monitoring videos, an image fusion method using pulse coupled neural networks in the contourlet domain is presented. This work is mainly for combining useful information of multiple monitoring sensors to gain a more comprehensive description of the scene. After contourlet decomposition, normalized contourlet coefficients of source images were inputted into the neurons of the pulse coupled neural networks. Through this process, the output pulses of these networks reflected the perception of human visual neural systems to detailed information of the scene. Since a stronger stimulus could make corresponding neuron fire earlier and fire for more times, both the first firing time and the average times of ignition in the linking neighborhood centered at corresponding neuron were selected as the salience measures for the high-frequency fusion rule. Experimental results prove the validity of our proposed method both visually and objectively.