Font Size: a A A

Nss With The Hvs-based Image Quality Assessment Method Study

Posted on:2010-04-18Degree:DoctorType:Dissertation
Country:ChinaCandidate:B LouFull Text:PDF
GTID:1118360272977769Subject:Circuits and Systems
Abstract/Summary:PDF Full Text Request
Digital image and video processing technologies have been developing rapidly during the last decade. Image quality assessment is fundamentally important in designing image processing algorithms, optimizing the performance of image systems and monitoring the quality of image transmission. Natural Scene Statistics (NSS) and Human Visual System (HVS), which describe characteristics of the object and subject in visual perception respectively, are used to analyze natural images and distortions. This dissertation is about automatic algorithms for quality assessment of digital images, which can predict image quality accurately, quickly and effectively, while full-reference, no-reference or reduced-reference image can be accessed. The major contents are as follows in general:1) Image quality pooling strategy based on perceptual property. Improves SSIM using weight map which get from front-end model of HVS. Considering visual masking, sensitive maps of error and texture are calculated respectively. Then sensitive maps are added to generate a perceptual importance map, weighted by this importance map, SSIM map is pooled into a value predicting image quality. This method emphasizes low-level property of HVS, which is ignored by SSIM. And the quality assessment performance is improved for images undergoing different kinds of distortions.2) Improves SSIM based on error segmentation. According the visual importance of different distortion regions, the distorted image is divided into three regions: texture region, dilating region and smooth region. The mean SSIM of each region is calculated and the dilated region is accounted for the region most sensitive to image quality. Define SSIM Distortion and model visual masking, SSIM of dilating region is compensated by SSIM of smooth region. Considering error segmentation and visual masking, the algorithm improves the quality assessment performance for all kinds of distortion images in image database LIVE2.3) Proposes a no-reference quality assessment method based on the power spectral model of natural images. The naturalness can be used for image quality assessment. The linear distribution of power spectral is an important law for natural image statistics. Meanwhile, image distortions change the shape of the power spectral, and the change is proportioned to the intensity of image distortion. In Contourlet domain, the powers of high level sub-bands are used to predict the powers of low level sub-bands. Pooling the distortion of powers of sub-bands, the qualities of the images distorted by JPEG2000 compression, Gaussian blur and Gaussian white noise can be effectively predicted. In addition, a new quality metric for JPEG compression is proposed. The proportion of the power on the border of image blocks and the power in the center of image blocks are used to measure the blocking effect. It validates that JPEG compression changes the spatial distribution of the image power.4) For reduced-reference quality assessment, the selection and transmission method of image quality sensitive attribute is developed. Different statistic in and between sub-bands are used as quality sensitive attribute for performance evaluation. Experiment results show that the variance of sub-band coefficients is short and the most efficient for quality assessment. Watermarking is a particular method for transmission of quality attribute, a new watermarking for image content authentication is proposed, in which moving regions are plotted out using motion detection as content protected region. According to different importance, the method provides hierarchical protection of image content. Experiment results show that the proposed watermarking is resistant to attacks such as partial tamper and MPEG compression.
Keywords/Search Tags:Image quality assessment, Natural Scene Statistics (NSS), Human Visual System (HVS), SSIM, Digital Watermarking, No reference image quality assessment
PDF Full Text Request
Related items