Improving sparsity of Convolutional Sparse Representation using a combination of a few dictionary filters

Convolutional Sparse Representation (CSR) [1] is a method of approximating image and audio signals. CSR represents signals as a convolution of dictionary filters and coefficient maps, and dictionary filters have partial features like filter of CNN. CSR has been applied in various fields, including CNNs and Distributed Compressed Video Sensing. In this paper, we propose to combine dictionary filters to improve sparsity and found that sparsity was improved.