Aiming at problem that block structure dictionary learning method based on sparse clustering

can not distinguishes the difference of numbers of the atomic support sets of the dictionary in compressed sensing. A method of learning the block structure dictionary by spherical K-means clustering is proposed

combine the dictionary atom support set into consideration

the dictionary atoms are clustered by cosine distance distinguish to form a dictionary with non-uniform block structure

and use block structure dictionary obtained by learning to reconstruct signal. The experimental results show that the matching between the dictionary and the image signal is better than that of discrete cosine transform（ DCT）

unstructured dictionary and block structure dictionary based on sparse clustering

it improves image reconstruction quality effectively and reduce signal reconstruction error.