Image Compression Based On Sparse Decomposition

Image compression is one of key steps in image processing and it has been widely applied to many fields of modern sciences and technologies. A lot of image compression methods have been proposed after many years study. And these methods, generally based on orthogonal transform, have good performance in many fields, especially for image compression at high bit rate. However, for image compression at low bit rate, these methods usually don't perform as good as at high bit. So it is necessary to develop a new method for image compression at low bit rate.Sparse decomposition becomes very popular in the study of signal processing in the recent years, because it can transform one signal into a sparse formation. Sparse decomposition is therefore applied to image processing quickly. In this thesis, the image compression is studied based on sparse decomposition and a new image coding scheme is proposed based on the result of image sparse decomposition.In this thesis, the image sparse decomposition is introduced firstly, and the characteristic and key problems of image sparse decomposition are mentioned. As the formation of image sparse decomposition result, sparse image presentation is introduced.In the following, matching pursuit, which is the most popular algorithm of image sparse decomposition, is introduced. Matching pursuit is much easier to understand and to apply than other algorithms, but it also has great computational burden. Therefore genetic algorithm is applied to MP-based image sparse decomposition. But the original genetic algorithm can't still afford the image decomposition, so, in order to have good reconstructed image quality and low computational burden, the genetic algorithm used in this thesis, is improved through many other methods.Based on the image sparse decomposition, the distribution of image sparse decomposition result is analyzed and an image coding scheme is then proposed based on the image sparse decomposition results. In this coding scheme, the order of results is adjusted according to the value of projection, which is the most important one of all the parameters of image sparse decomposition, and the projection is pre-processed by pre-difference. After that, the range of projection value decreased largely. Because the range of projection value is much larger than those of other parameters, the wholeimage coding efficiency can be improved if the efficiency of projection is improved. So the image coding scheme proposed in this thesis can improve the whole image coding efficiency largely.At the end, the experimental results show that the image coding scheme proposed in this thesis can decrease the coding redundancy of projection, and improve the efficiency of image coding. Experimental results also show that, at low bit rate, the newly presented method outperforms many other image compression methods in the objective and subjective qualities of decoded images.