| With the constant development of information industry, digital image inpainting technology has become increasingly important in the field of digital image processing. The principle of image inpainting technique is to use information known to approximate the unknown image region. In recent years, image processing technology based on sparse representation has been widespread concerned. This thesis is based on the theory of sparse representation to sparse the image signal firstly, and then reconstruct the original image using a reconstruction algorithm. The main research work of this thesis is as follows:1. The theory on sparse representation is introduced. And the one-dimensional data is tested to verify the effect of data recovery and the effects of different parameters on the signal inpainting are studied through some comparative experiments. Also, we achieve the image inpainting based on sparse representation is simulated through MATLAB platform.2. The adaptive learning dictionary using K-SVD algorithm compared to general discrete cosine transform(DCT) dictionary is researched, which could enhance the adaptability of dictionary to different images. Combining the K-SVD algorithm and the sparse representation algorithm, the different types of damaged images are tested. The results show that the improved algorithm has better repair effect and adaptability. For different types of images, the learning dictionary has better repair effect.3. We improve the image restoration algorithm by extracting the image pixels in fully overlapping ways instead of processing non-overlapped patches and using adaptived MOD dictionary for image inpainting. To overcome the problem that the K-SVD algorithm is so complicated and the coefficient matrix of MOD algorithm may be irreversible, we present the revised version of MOD algorithm and apply it to the fully overlapped image restoration algorithm. Optimized MOD algorithm has lower complexity and higher inpainting effect than K-SVD algorithm. |
PDF Full Text Request