The Research On Objective Quality Assessment Of Natural Images

The objective quality assessment of natural images and videos is the main work in the thesis. From the viewpoint of science instead of art, the quality of images or videos is predicted by objective methods, where the result is not given by subjective judgments. The image and video quality plays an important role in the accuracy, adequacy and effectiveness of the information what they convey. In the fact, images and videos are subject to many kinds of distortions during acquisition, processing, coding, transmission, recording and so on. The aim of researching on objective image and video quality assessment is to find automatic and accurate models or methods to substitute for the subjective evaluation. The main work and contributions of the thesis are summarized as follows:(1). The procedure of image an video quality assessment is analyzed from the viewpoint of information evaluation, that is to say, images and videos, which belong to vision information, are added in the procedure of information evaluation activity, where the consult is that image and video quality assessment is one kind of so many evaluation activities. Some concepts in connection with quality are discussed. Objective quality assessment methods and key problems are extensively reviewed, including full reference, reduced reference and no reference. The relations among models and conformity patterns are introduced briefly. An experiment platform which is based on VC++6.0 is given and is the foundation work of the whole project.(2). Objective quality assessment of natural images is investigated. First, the image properties are discussed in the task of quality assessment. Second, based on self-similarity and sparseness of natural images, two kinds of full reference quality assessment models are proposed with multifractals and independent component analysis respectively. Experiment results show that these two methods could get satisfied prediction accuracy in accordance with subjective judgments. Third, the image complexity under the quality assessment task is researched. The role of local variance in the image quality assessment is studied. Then the relationship between image complexity, which is described by ICA characteristics, and subjective judgments and objective predictions are discussed respectively.(3). Color information in the image quality assessment is explored. First, the work of color image quality assessment is surveyed, including the choice of color spaces, the extraction and expression of color information. Second, the influence of color spaces on quality assessment methods is indicated by evaluating an objective quality assessment model in different color spaces. Third, SHSIM and SAC&SMC are proposed, which combine structure and color information of images. Experiments show that the join of color information could improve the accuracy of objective quality assessment models.(4). No reference video quality assessment is reviewed. An objective quality model with space-time redundancy information in the videos is proposed, which is tested in VQEG-Phase I subjective database. Experimental results show that the model could get comparatively satisfied prediction, which is in accordance with subjective judgments basically.