Research On Full Reference Image Qaulity Assessment Algorithm

With the rapid development of multimedia technology, digital image has been more and more widely used in various fields. But during the acquisition, compression and transmission process of digital images, the distortion of images has also been taken in. It is increasingly important to evaluate and modify digital image processing systems by making proper image assessment algorithms. As the most reliable of image quality assessment, the full reference one was focused on in this paper. With their high convenience and accuracy, Structural Similarity(SSIM) and Gradient Magnitude Similarity Deviation(GMSD)has been widely used to assess image quality. This paper mainly focus on the 2 algorithm and overcome their defect. The main contributions of author are outlined as follows:1 White noise distortion of the image and non-white noise white noise distortion of the image has been distinguished. Since most image processing standard databases provided are in accordance with the type of distortion of the image, and the distortion is also the source of our assessment of image quality. The five types of image distortion described in LIVE image database II have been studied, which was combined with the characteristics of the type of image distortion and image degradation model. In this way, a method of distinguishing white noise distorted images and non-white noise distorted images has also been proposed in this paper. The difference image between distorted image and reference image was used to distinguish white noise distortion and non-white noise distortion of images.2 A method of image quality assessment based on type of distortion has been proposed. On the basis of distinguishing the type of image distortion, a method of image quality assessment processing the image depending on its distortion has been presented. For non-white noise distorted images, improved SSIM is used to assess them. In this case, the reference image was the original reference image superimposed on its edge-image, and the distorted image was the original distorted image superimposed on its edge-image. Experimental results on 982 images of LIVE image database II of five types of distortion showed that the proposed method had a better consistency with subjective quality assessment, a higher stability and accuracy.3 A method of image flat background region suppression has been proposed. In order to overcome the complexity and interaction of existing image segmentation algorithms, a suppression gradient image was conducted with a simple morphological adaptive threshold operation, combined with morphological closing operation to fill small voids of the gradient suppressed image. Then the and operation was applied to get the gray-scale flat background region suppressed image.4 A method of image quality assessment based on flat background region suppression was proposed. After segmentation of the image, a flat background region suppression image was achieved. Then the we needed to get the Gradient Image, then Gradient Magnitude Similarity, Gradient Magnitude Similarity Mean, and Gradient Magnitude Similarity Deviation as the final score of image quality assessment. Also, experimental results on LIVE 982 images of five types of distortion showed that the proposed method had a better consistency with subjective quality assessment, a higher stability and accuracy.