Improving Fusion of Surveillance Images in Sensor Networks Using Independent Component Analysis

In this paper we present a novel algorithm for fusion of multimodal surveillance images, based on ICA, which has an improved performance over sensor networks. Improvements have been demonstrated through separate training process for different modalities and the use of a fusion metric to maximise the quality of the fused image. Sparse coding of the coefficients in ICA domain is used to minimize noise transferred from input images into the fused output. Experimental results confirm that the proposed method outperforms other state-of-the-art methods in the sensor network environment, characterized by JPEG 2000 compression and data packetization.