基于相似性的图像质量评价算法研究

发布时间：2018-06-17 00:27

本文选题：全参考图像质量评价 + 区域级　；参考：《北京邮电大学》2016年硕士论文

【摘要】：随着科学技术的发展以及互联网的普及,数码技术得到了广泛的应用,然而数字图像的获取、处理、编码、存储、传输和重建这每一个必要的环节都会影响图像的质量,而图像质量的下降势必影响到图像所表达信息的充分性与准确性,所以图像质量评价显得尤为重要。图像处理办法在不断改善,高动态范围图像的应用也越来越广,图像评价办法也需要不断改进以适应新需求。优秀的数字图像评价方法不仅能够准确地反映图像的质量,而且能对图像处理技术提供良好的改进方向。本文围绕全参考图像质量评价主题即失真图像所对应的参考图像可以获得,在学习和研究已有基于相似性图像质量评价的基础上提出了新的方法。首先,本文在已有一级加权策略基础上提出了二级加权方法,对已有的图像质量评价算法进行了改进。已有的加权策略都是集中在像素级的加权上,而人眼感兴趣的往往是某一块区域,而不是某一个像素点。区域加权算法具体为:区域内对像素级局部质量值进行加权得到各个区域的质量值,区域间通过各个区域的权重对区域间的质量值进行加权。其次,人眼是一个很好的信息提取者。图像若有失真,人眼从图像中感受的内容信息就会改变。本文基于人眼的信息提取特性对图像的内容信息进行建模,并结合梯度和色度特征,提出了一种基于内容信息相似性的图像质量评价算法。第三,本文融合修正的梯度相似性和图像的自然性实现了对色调映射图像的质量评价。在图像质量评价方面公开的数据库上进行的试验表明,本文提出的三种算法均取得了良好的性能。
[Abstract]:With the development of science and technology and the popularity of the Internet, digital technology has been widely used. However, the acquisition, processing, coding, storage, transmission and reconstruction of digital images will affect the quality of images. The degradation of image quality is bound to affect the adequacy and accuracy of the information expressed in the image, so it is particularly important to evaluate the image quality. The method of image processing is improving and the application of high dynamic range image is becoming wider and wider. The method of image evaluation also needs to be improved to meet the new needs. Excellent digital image evaluation method can not only accurately reflect the image quality, but also provide a good direction of improvement for image processing technology. In this paper, the reference images corresponding to distorted images can be obtained around the subject of full reference image quality evaluation, and a new method based on similarity image quality evaluation is proposed in this paper. Firstly, based on the existing first-order weighting strategy, a two-level weighting method is proposed to improve the existing image quality evaluation algorithm. The existing weighting strategies are focused on the pixel weighting, and the human eye is always interested in a certain region, not a pixel point. The region weighting algorithm is: the local quality value of each region is weighted in the region, and the quality value of each region is weighted by the weight of each region among the regions. Secondly, the human eye is a good information extractor. If the image is distorted, the content information that the human eye feels from the image will change. In this paper, based on the information extraction characteristics of human eyes, the content information of the image is modeled, and an image quality evaluation algorithm based on the similarity of the content information is proposed by combining the gradient and chromaticity features. Thirdly, this paper combines the modified gradient similarity and the naturalness of the image to evaluate the quality of the hue map image. The experiments on the database of image quality evaluation show that the three algorithms presented in this paper have good performance.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】