Research Of Super-resolution Reconstruction Based On Sparse Representation With Multi-component Dictionary

Super-resolution reconstruction based on sparse representation has been paid more attention in recent years, which need restore the high-frequency information from the low frequency data existing in the low-resolution image. Therefore, it is the key to construct the dictionary containing the high frequency part for super-resolution reconstruction.Every image can be decomposed into texture and cartoon using morphological component analysis (MCA) where these parts are both mutually independent. Then, we train two over complete dictionaries, one of which is based on texture, the other only contains cartoon.The method proposed in this paper first decomposes training images into texture and cartoon. And then sample these two sub-graphs according to a certain proportion, after that an interpolation method is used to enlarge the size of texture (or cartoon) sub-graph as big as the training image. Then we can get the high frequency information of texture (or cartoon) sub-graph accordion to the subtraction between original image and the enlarged one. Last but not least, we need learn two dictionaries, one of which contains high frequency information obtained from the high frequency data, the other one can be achieved from the enlarged image’s blocks.For a testing image, first decompose it into texture and cartoon as the training images. And then enlarge texture part (or cartoon part) as big as the target size using the same method in the training step. Thus, we have gotten the low frequency parts which belong to the high resolution image. We can construct the high frequency part using high frequency dictionary with sparse representation vector solved by low frequency dictionary. The texture (or cartoon) belongs to high resolution is superposition of construct high frequency content and enlarged low frequency data. Finally, the target high resolution image is obtained after the texture and cartoon superimposed. Experimental results show that reconstruction effect can be improved using these two components dictionaries. Synthesis model using sparse representation has become a hot research scholars present, but the analysis model sparks scholars’ interest in the recent years. In this paper, I have focused on three points. The first one is what are analysis signal and the differences between synthesis model and analysis model. The second one is how to construct a dictionary in analysis model. The last one is achieving image super-resolution reconstruction with Analysis K-SVD algorithm. Experimental results show that Analysis K-SVD is doing better at some image super-resolution reconstruction than Synthesis K-SVD.