Applications Of Natural Image Statistics In Image Processing

Recent years, with the rapid development of the computer technology, there exists a great demand of multimedia technology. As a result, the image/video processing techniques are significantly promoted. In image processing field, researchers are most interested in the so called natural images. Typically, natural images are referred to those images that are adapted to the animal visual system, which form a tiny set in the image space. Despite the fact that natural images are highly complex, researchers have found a few inherent laws of its statistics.Natural images are information redundant, which is the foundation of image compression. Principle Component Analysis (PCA) is a general de-correlation technique of data processing, and has incorporated into data compression. We proposed a new image compression algorithm based on PCA. To further improve the performance of our algorithm, we made it adaptive according to the content of the image. Experiments show the proposed scheme archives better PSNR compared with JPEG at low compression ratio, and its compression performance is significantly improved compared with traditional PCA method.Image Quality Assessment is the new field of image processing. Among all types of image quality assessment, the no-reference is the most difficult. Natural image statistics is considered to be the most promising one to solve no-reference image quality assessment problem. Independent Component Analysis (ICA) has drawn great attention in recent years, and has been widely applied in signal processing. We proposed a new no-reference image quality assessment algorithm based on ICA. We utilize ICA to model natural images and extract features, and measure the quality of images by comparing its features with those extracted from distortion free images. Experiment results show the proposed algorithm performs well on different types of distortions. Besides, we proposed the concept of image local correlation. Based on the fact that blur will increase the local correlation of images, we proposed a new no-reference blur metric. Image quality is measured by comparing the variation of local correlation. When tested on different image databases, the proposed no-reference blur metric achieves good performance that comparable to some state-on-the-art methods and has high robustness.