A Study Of Impulse Noise Removal Based On Structural Feature

As the important medium of information transmission, images are always contaminated by noise when they are captured, transported and stored. This affects our understanding to image, which makes noise removal has gained wide attention. In recent years, with the continuous development of technology, there is a great demand on high quality and high resolution image in national security, scientific research and daily life. Therefore, how to recover uncontaminated images from corrupted images has always been a hotspot issue.There exist two characteristics in corrupted images: The first one is that only a portion of pixels are noised while others are noise-free, which means that we can restore uncontaminated images by the information of the noise-free pixels. The second is that the intensity value of noised pixels exists at the two ends of the histogram, which means that it is possible to detect noised pixels from noised images according to the detection algorithms.The procedure of the existing denoising algorithms includes: detecting impulse noise from corrupted image; using noise-free pixels within a local window to calculate the centered noised pixel in a statistical way; traversing the whole noised image using sliding window to restore clear image. These methods have some shortcomings.First, as we all know, the removal performance of a method largely depends on its description ability to local structural features of an image. In the calculation process of the intensity value of noised pixels, existing methods make full use of the statistical features while ignoring the structural features.Second, there exists correlation between image pixels. And some applications in image processing have proved that we can get better results when the correlation is taken into consideration. Existing methods consider that the centered pixel is individual when estimating its intensity value within local window rather than combine with the correlation and lack the process of jointly solving pixels in neighborhood which makes it is difficult to improve the restoration precision.Finally, impulse noise detection is the first step in the impulse noise removal process. Only if damaged pixels are detected, can they be denoised. By this token, the higher detection accuracy is, the better recovery performance is. Therefore, the existing methods always spend a lot computational expense in designing high precision and accuracy noise detection methods.Aiming at those problems, this paper proposes an impulse noise removal method based on structural features. The key distinction between the proposed method and existing methods is that this method analyzes and applies both statistical feature and structural feature which is achieved by autoregressive modeling to realize effective denoising. In addition, this method process a group of pixels at a time jointly rather than estimating the individual pixels in isolation via solving the optimization during the process of autoregressive modeling and image recovering. In the end, owing to jointly estimating noised pixels, some false alert rate would have little effect on recovery performance. The computation resources have been spent on noise suppression. Therefore, this method adopts simple detection based on histogram to detect impulse noise. In this method, all noise-free pixels are used as the known conditions to supervise the denoised procedure for higher recovery performance and better detail information preserving. The experimental results demonstrate that the proposed method produces better results than existing denoising methods both subjectively and objectively.