SISCHMAX: Discovering common contour patterns

Discovering common shape contour is a promising topic. However, local position and scale variance always leads to the mismatch of similar contours. In this study, we propose Shift Invariant Sparse Coding HMAX (SISCHMAX)to address this problem. Shift Invariant Sparse Coding is used to learn the configuration of line responses on the output of HMAX C1 layer. And we test the proposed method on Caltech101 dataset for discovering object contour. Due to the tolerance of local shift of the lines, sparse coding extracts better object contours using HMAX C1 outputs instead of gray value.