Abstarct: One of the most important requirements of image processing applications is efficient methods for image quality assessment. Such methods are expected to predict the subjective scores resulted from the human judgments. But the objective scores provided by the existing measures are not adequately consistent with the human subjective scores. In addition, many of them are designed especially for particular distortions and have limited popularity. Moreover, most of the outstanding measures suffer from the high computational complexity and hence lose their effectiveness in online applications. In this dissertation, we propose a novel frxamework for image quality assessing, in which the image quality is evaluated baxsed on the human judgment process. This frxamework with the aim of modeling the human visual perception process includes three major stages as follows: 1) estimating the illumination (brightness or darkness) and contrast of the image, 2) evaluating the possibility of image content detectability, and 3) assessing the possibility of image details perception. What distinguishes the proposed frxamework from the existing ones is its independency from the image distortion type. Indeed, here we try to assess the amount of success in perception, in any of the above three steps, instead of gauging the amount of image distortions. In the first stage, we propose the novel concept of Eigen-histograms, originated from the image local histograms, and employ them in a learning system to build a no-reference contrast evaluator. In the second stage, we found that success in the perception of the image content is strongly related to the quality of basic structural components of the image, such as edges, and also the absence of spurious contents like additive noises. To assess the quality of the structural components of the image, we employ the non-negative matrix decomposition, which is sensitive to the constituent components of input matrix (i.e. primitive parts of image). The region smoothness modality modification is the most important change resulted from the additive noises. To assess this kind of distortions we propose a novel image smoothness measure employing the maximally stable extremal regions and use it to construct an efficient quality evaluator. In the third phase, structural components and non-structural details are separated using the singular value decomposition, and employed in a measure to assess the image details. In the aggregation step, as the last ones, we made suitable choices from the above criteria, so that the final solution has reasonable computational complexity, and simultaneously, the aggregation errors are minimal. Experiments conducted on popular image databaxses (LIVE, CSIQ, TID2008 and TID2013) indicate that the proposed measure is highly consistent with the human perception and comparable to the state-of-the-art IQA methods in terms of prediction accuracy, generalization ability and computational complexity. Keywords: