Font Size: a A A

Image Quality Assessment Based On Visual Attention Mechanism

Posted on:2017-04-23Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z S ChenFull Text:PDF
GTID:1108330491464157Subject:Physical Electronics
Abstract/Summary:PDF Full Text Request
As the requirement of people for the image quality becomes higher and higher, the image quality assessment (IQA) plays a more and more important role.The IQA methods can be categorized as objective measures and subjective measures. The results of subjective assessment could perfectly deliver viewers’ feelings. But it’s time consuming, costly and inconvenient. The objective IQA is relatively low cost and convenient, it has been regarded as the vital one in IQA methods. However, the results of objective measures do not always correlate well with perceived image quality. In order to improve the consistency of subjective and objective evaluation results, a novel visual attention(VA) computational model had been introduced into the objective IQA in this thesis, the experimental results matched with subjective assessment well.For the existing bottom-up VA computational model, the final visual saliency map (VSM) is obtained through mean or sum of the saliencies of all visual features. However, VA of VS for different visual features is not identical according to the selective visual mechanism. In this thesis, a new fusion strategy weighted by the saliency of each visual feature was submitted, which can quantitatively describe the selective visual mechanism. The VS model in this thesis firstly extracted visual saliencies of all lower-level features and calculated the contribution ability of each saliency, and then neglected the distracting informations based on an apriori threshold and then combined the effective informations to construct the final VS. Experimental results show that the performance of the proposed VS model can effectively suppress clutter targets, highlight contour, and reduce the misjudgment rate. The proposed VS model, considering the local and global saliency, is more conform to the human visual system(HVS).The different image regions, with the different visual importance, give the people different subjective perception. The VS of the image plays an important role in IQA. A new grayscale IQA algorithm, combining VS with structure similarity has been proposed. The image quality was calculated by weighting distortion map based on pixel VS. That is, the different regions of interest were assigned to different weights. Experimental results show that the proposed grayscal IQA algorithm, with better accuracy and lower complexity, had a better consistency with subjective assessment results. The results also show that the proposed grayscale IQA algorithm had obvious advantage in fast fading images and gaussian blur images, but was not good at the JPEG2000 compression images. Based on the proposed grayscale IQA method and the color visual characteristic in visual preception, a new color IQA method considering color distortion was proposed. The proposed color IQA method obtained the final image quality via integrating the structure similarity of lightness, chroma and gradient magnitude in YUV color space to get the image distortion index, and weighting the index based on pixel visual saliency. Experiment results show that the proposed color IQA method had higher computational accuracy than the other IQA methods under a moderate computational complexity, especially for two types of distortion images, such as fast-fading and white noise. But it did not have obvious advantage in JPEG2000 compress images. The results also demonstrate that performance of our weighted-VS IQA metrics is superior to the performance of no weighted-VS IQA metrics.
Keywords/Search Tags:selective attention mechanism, visual attention, integration strategy, visual saliency, image quality assessment
PDF Full Text Request
Related items