Research On No-Reference Image Quality Assessment

In recent years, as mobile Internet develops rapidly, image and video information has become a major data traffic of the Internet. With the constant perfection of the network basis instrument, as well as the popularity of Mobile3G Network, users’demand for visual information increases continuously. As a result, how to guarantee the quality of the end-user experience becomes the key problem. This makes image and video quality assessment which is closely related to the quality of the user experience presents unprecedented importance.The type of the image’s distortion which is introduced during acquisition, processing, transmission and storage process is uncertain. Moreover, little information of the original image can be available in practical applications. Therefore, research on universal no-reference image quality assessment is proved more valuable for practical applications. By the study of universal no-reference image quality assessment, this paper puts forward the following research results:1) Classified the existing universal no-reference image quality assessment algorithms and analyzed the key factors to considerate when designing universal no-reference algorithms, which laid the foundation of the subsequent research.2) Proposed a no-reference image quality assessment algorithm based on salient region. In order to make evaluation of the quality of distorted image, this algorithm measures the deviation of distorted image and original image in spatial natural scenes statistic. According to the characteristics of image quality assessment, this paper designed a salient region extraction algorithm which is applicable to image quality assessment. Introducing the influence of salient region improves the overall performance of the algorithm, which achieves better subjective consistency.3) Proposed a no-reference image quality assessment algorithm based on high-level semantic. This algorithm breaks the "semantic gap" between the low-level image feature and image quality through the introduction of high-level semantic. Using probabilistic latent semantic analysis technique, it compared the distribution difference between the reference image and the tested image on the latent semantic space, thus assessing the quality of the tested image. The experimental results show that the objective evaluation result of this algorithm is consistent with human subjective perception.