Image Reconstruction Based On Smoothed L0

Image reconstruction is a hot point in signal processing industry, which use the low frequencies by image data acquisition equipment to recover the high frequencies of the image. The existing image devices including electronic microscopy imaging system, X-ray and infrared image system, optical imaging system, and their actively or passive imaging processing will degrade the image quantity. The image data transmission and restoration also will lower its visual effect. In daily life, medicals need high digital image to help doctors analysis the lesion and accurately capture etiology. Digital scanning devices need high digital image to give the users a complete information details. Mobile phones, cameras, computers and other digital products need high definition video to help manufacturers obtain the best users’ experience. So scientific research and engineering applications of digital imaging systems have become increasingly demanding, and more and more people pursuit high-quality, high-resolution images. A simple continuous improvement depends on the hardware equipment, will cause the system to become more and more complex, the cost will increase substantially. The image super-resolution technology is the one just use the appropriate signal processing technology, and no need to change the existing conditions of the physical device which you can get a high-resolution high-quality images to meet the needs of the people. It has a large advantage both in technology and cost. So it is widely used in high-definition digital television (HDTV), military surveillance, public safety and health impact of other fields.As machine learning and pattern recognition technology in recent years have been paid more and more attention, image super-resolution technology based on learning has also been an unprecedented development. The relatively sparse representation which presented to be a hot research direction over the past decade, has been well applied in image super-resolution technology and image denoising. This paper study the researches of super-resolution image reconstruction techniques, and combine with application of the sparse representation in image reconstruction problems. Then we also gave improvements in the math model, and proposed super-resolution image reconstruction based on smoothed L0 model technology, which used in the image restoration and image denoising. The main work and innovation points is presented by:1) Deeply studied the sparse representation theory and the classical image super-resolution technology, and using smoothed L0 representation theory to approximation the sparse coefficients which are solved the problem in the sparse representation theory. Smoothed L0 approximation replaces the L1 norm optimization problem which should be a strict condition. When the sample library are learned by the smoothed L0 approximation theory, we will get an approximated sparse dictionary. For its smoothness of the smoothed L0 model, the gray concentrated infrared images have better reconstruction results.2) Analyzing the smoothness of smoothed L0 model and the image denoising theory via sparse and redundant representations over learned dictionaries carefully, we gave a solution of the sparse representation and obtained a good performance of image denoising by smoothed L0 approximation model through multiple sets of test experiments.