Abstract:

Aiming at the deficiency of existed algorithm with single dictionary and redundant coding, the dictionary combined convolutional sparse coding and K-SVD is constructed in terms of the hierarchical properties of human visual perception systems and the lateral inhibition and competition mechanism of neurons. Furthermore, an effective algorithm based on convolutional matching pursuit and orthogonal matching pursuit is proposed to obtain sparse image representation with the combined dictionary. The experimental results indicate that the combined dictionary can adaptively match up image geometric structures such as edge, blob, texture. In comparision with convolutional dictionary and redundancy dictionary based on K-SVD, the combined dictionary has sparser image representation.