It is known that in practical applications, the quality of the images is often degradedduring the process of achieving, storing and transporting etc. So a systematic image de-noising processing is very necessary.In the recent years, motivated by sparse representation of the signals, a novel efficientimage de-noising method based on over-complete dictionary generated by data, that is K-SVD algorithm suggested by Elad, Aharon and Bruckstein. In this paper we propose anovel dictionary design algorithm, a new method dictionary learning method based oncontent-based clustering and K-SVD algorithm. Firstly,the image can be divided into Mclasses with K-means algorithm based on the geometric content of the image. Then forevery sample image clusters we use the K-SVD algorithm to train dictionaries singly. Atlast an over-complete dictionary is generated by combining these dictionaries in order toachieve the sparse representations of the image and the purpose of image de-noising. Iwill apply this algorithm in the testing images.We can see that the algorithm proposed inthis paper has a better de-noising results comparing with the K-SVD method. |