Abstract:

In order to solve the problem of decreased image quality due to the poor robustness against noise in compressed sensing (CS) reconstruction, the CS model based on adaptive learning is studied and a new method applied to image reconstruction based on the blind compressed sensing (BCS) model is proposed. The proposed method employs the idea of alternating update of sparse base and sparse matrix from the BCS model, and adopts the strategy which combines the image redundant transform and initial combination of discrete cosine transform base, thus solving the problem that the sparse base is hard to express, restraining the noise and improving the image construction quality. The experimental results show that the proposed method has better robustness against noise and maintains better image texture information than the orthogonal matching pursuit based on wavelet method and the total variation method.