Efficient surveillanve video compression using dictionary/transform learning

Abstract

Techniques used for sparsifying the signal in dictionary or transform domain are quite popular in signal processing. In this work, we propose a method for e cient compression of surveillance videos. We observe that in these videos background remains xed and only foreground moves. In order to take advantage of this scenario, we use Multi-Layer Convolutional Sparse Representation (ML-CSC) to sparsify the background of surveillance videos and reconstruct the background and add the subtracted foreground at decoder side. Our work also shows the e ectiveness of dictionary learning in compressing the light- eld data. We will also use a new algorithm other than dictionary learning which is transform learning which is more robust and fast compared to dictionary learning, and will compare the two algorithms on the basis of compression ratio.