Research On Image Restoration

The visual is one of the important forms of the human experience. Over the ages scientists and engineers have strived to capture precious and high quality images. Recently, the popularity of digital cameras improved the quality of the image acquisition and generated a lot of images and photos. However, it is relatively difficult to obtain high quality images even for the experienced photographer in some cases. In addition, images can be degraded during the process of print, transmission and scanning due to the halftoning process and channel noise. When using a scanner to scan the printed image, there are a lot of screen patterns and noises in the scanned images. In the practice application, we should adopt inverse halftoning and denoising algorithm to recover images. Therefore, the method of image restoration for inverse halftoning and denoising, is the basis and core technique of image processing. Many image processing applications are built on the quality of image restoration.Image restoration is a challenge problem. Inverse halftoning and image denoising problem are ill-posed, where the number of unknown values outweights the number of observations. As a result, it is necessary to adopt image content to improve the restoration process. Recent Research focus on using content based adaptive method to improve image restoration quality. In this work, we introduced image content analysis approach to address the image quality restoration. The contributions of the paper mainly include:1. A halftone image analysis approach based on discrete Voronoi diagram is proposed. With the image geometry presentation, Voronoi diagram was introduced to halftone image local features analysis. The image feature including tone values, neighborhood constraints and gradient can be calculated from above analysis approach.2. We proposed an inverse halftoning algorithm based on discrete Sibson interpolation. Firstly, the discrete Voronoi diagram is introduced to estimate halftone dot's gray value. Then, an improved discrete Sibson interpolation approach is used to calculate other pixels gray value. Compared with other inverse halftoning algorithm, our algorithm has better applicability.3. We extended the Voronoi diagram analysis approach and proposed an adaptive energy diffusion algorithm for inverse halftoning. We treated the inverse halftoning process as Gaussian energy diffusion and introduced more images content features to this process. Compared with the discrete Sibson interpolation algorithm, we can get more details and smoother images.4. An image statistic based NL-means image denoising algorithm is proposed. The image statistic method based on Weibull distribution is applied to image patch content analysis. According to the content analysis, image patches are classified into three types:smooth type, edge type and texture type. For different type of patch, we introduced different similarity method and different patch parameters. Based on the results from various different images, our content-based NL-means algorithm is shown to have better performance in both PSNR and visual quality, comparing with the traditional NL-means algorithm.A scanned image pre-process system has been designed and implemented based on the proposed algorithms. Meanwhile, the pre-process approach has been introduced into textile product CAD system.