Research On Color Image Reconstruction Method Based On Compressed Sensing Theory

With the continuous progression of the information technology,color images have played more and more important roles in the fields as biomedical science,aerospace industry,remote sensing measurement and communication engineering etc.However,in the process of image acquisition,it is inevitable to face the problem of noise mixing, image distortion, lack of content and so on.Therefore,it is significant to study the technology of color image reconstruction.And as image quality improves continuously,traditional image processing technology can hardly meet the demands of modern age because of limited by the Nyquist sampling theorem.Then a new theory called of Compressed Sensing,owing to reconstructing signals with its own sparsity but not the frequency,can restore the original signal accurately with a little amount of the sampling data. Based on this characteristic, the Compressed Sensing theory has obvious advantages in the field of digital image processing.In this paper,we take the Compressed Sensing theory as theoretical basis,and apply the theory into the researches of the color image denoising and super resolution reconstruction.Firstly,elaborate and analyze the Compressed Sensing theory by using mathematical technique,in which-according to the demands of specific experimentswe place emphasis on how to make the original signal sparse representation and the conditions to be satisfied,to lay the theoretical foundation for the experiment.And in the specific research,when facing a low resolution input image within noise.we need to denoise the input image,to get a pure low resolution input image without noise,and then take super resolution reconstruction on this image,to obtain the final high resolution output image.Therefore,all the experiments can be divided into the following two parts.In the experiment of color image denoising,aiming at the problems of long processing time cost and the low restoration image quality existing in traditionaldenoising algorithms,in this paper,we research on the improved K-SVD algorithm.We improve the K-SVD-based grayscale images denoising algorithm,and apply it into color image denoising researches,to avoid the possible artifacts,and then extend this improved algorithm to process images destroyed by strong noise and irrelevant texture,to accomplish the color image inpainting task.As the experimental results show that,compared with traditional principal component analysis algorithm,the improved K-SVD algorithm based on the sparse representation has obvious advantages on both the restoring image quality and processing time.In the experiment of image super resolution reconstruction,aiming at the problem of low super resolution accuracy existing in traditional interpolation algorithms,in this paper,we research on the reconstructing algorithm based on dictionary learning,according to the sparsity in the image itself,co-training two dictionaries of low resolution image patch and the high one,so to generate the high resolution output image with the two dictionaries.As the experimental results show that,compared with traditional traditional interpolation algorithms,the algorithm proposed in this paper has obvious advantages on the output image quality,meanwhile,this algorithm has the high robustness to the noise in the input image,which traditional interpolation algorithms do not.