Classification with Dictionary Filters in Convolutional Sparse Representation

Focusing on Convolutional Sparse Representation (CSR), this paper proposes a classification method using multiple convolutional filters obtained from images. Subspace Method (SM) projects a single input image onto a linear subspace representing image features and classifies it into the corresponding class. Another application of the SM is the Mutual Subspace Method (MSM), which uses subspaces of multiple images as input. In the proposed method, multiple convolution filters are obtained from CSR to generate subspaces, and classifications are performed by MSM to improve the recognition accuracy compared to SM.