Image Quality Assessment Based On Natural Computation And Visual Attention

In recent years, the scientific theories and technical means of digital images have developed rapidly, and have been widely applied to such fields as communications, industry, medicine, remote sensing, military, etc. They promote the development of human scientific research, the gains of social productivity and the improvement of life style greatly. However, during the process of acquisition, processing, transmission and recording, images will inevitably suffer from distortion or degradation to some extent because of the imperfections of imaging systems, processing methods, transmission media, and recording equipments, which brings great inconvenience in studying and solving problems, and understanding the objective world. Consequently, it is of great significance to evaluate the quality of images objectively, effectively and reasonably.As the terminal for receiving visual information, the human visual system is the most reliable instrument to assess the quality of images. It will improve the performance of evaluating algorithms definitely if we take the characters of the human visual perception into account. So based on the perceptual features and information processing mechanism of the human visual system, some methods have been given in this paper: 1) Combining the perceptual characteristics of visual system, including multi-channel and contrast sensitivity, a natural computation based full reference image quality evaluation method in wavelet domain is proposed, which uses the wavelet decomposition to simulate the multi-channel characteristic of human visual system and utilizes natural computation to obtain the scale factor(i.e. the contrast sensitivity factor of each subband). Experimental results show that the proposed method has better consistency with the subjective mean opinion score (MOS). 2) According to that the human eyes are more sensitive to structural information, an improved version of structural similarity (SSIM) algorithm is presented, which differently treats the images'forming factors, including luminance, contrast and structure, and captures the sensitivity coefficient of the structural factor by natural computation to highlight the importance of structural information. 3) Using visual attention mechanism to acquire the attentive regions of an image selectively, a type of attentive regions based pooling perceptual error image quality evaluation method is developed, which differently handles the image regions with different saliency. It can be used as a weighting form to optimize the existing image quality evaluation algorithms which pool perceptual error in pixel domain. 4) A reduced reference image quality assessment metric based on visual attentive features is proposed, which computes the differences of intensity, color and orientation between the original and the distorted images respectively, and furthermore obtains the sensitivity factors of the three features(i.e. intensity, color and orientation) corresponding to different distortion types with natural computation. Experimental results demonstrate that the proposed method has better consistency with the subjective MOS, which can accurately reflect the human eyes'subjective feelings for the image quality.