Color Image Super-Resolution Based On Inter-Channel Correlation

In recent years,single-image super-resolution technology has been widely used in the field of image processing,such as image enhancement and restoration,medical imaging processing,intelligent monitoring and processing,and so on.The task of single image super-resolution is to reconstruct the corresponding high-resolution image from a given low-resolution image.Although many super-resolution methods have been proposed,most of them are designed to improve the resolution of a single channel.When dealing with color images,a typical solution is to separately process each color channel.Another solution is to convert low-resolution images from the RGB space to the YCbCr space.Due to the reason that human eyes are more sensitive to the luminance component than the chrominance component,it is reasonable to apply the sophisticated super-resolution methods only on luminance channels.For the remaining chrominance channels,only the simple Bicubic interpolation is used.In the above two solutions,the correlation information between color channels is not fully utilized.Therefore,false color or artifacts might exist in the reconstructed images.In order to efficiently integrate the inter-channel correlation information of color image into the framework of single-image super-resolution method,two inter-channel correlation-based color image super-resolution methods are proposed in this thesis:(1)In order to utilize the inter-channel correlation information effectively,a super-resolution algorithm based on inter-channel correlation is proposed.In the proposed model,in order to effectively make use of the correlation among the color channels,the total variation(TV)model is used to characterize the structural characteristics of the correlation among the channels.Compared with the existing multi-channel constrained model,two improvements are made:firstly,the L1 norm,rather than the L2 norm,is applied to measure the correlation among three color channels.Secondly,in addition to the correlation information among the color channels,the average signal of the three color channels is also considered.(2)In the above algorithm,although the inter-channel correlation is used,non-local self-similarity of images is not utilized at all.In order to improve the performance of the algorithm more effectively,the constraint of non-local self-similarity of color image is further integrated into our proposed model,so that the prior information in images can be well utilized.Through extensive experiments,it is demonstrated that the proposed algorithm is superior to many existing methods in terms of the subjective quality and the objective quality.