基于人类视觉系统的屏幕图像质量评价方法研究

发布时间：2018-02-24 16:23

本文关键词： 图像质量评价人类视觉系统屏幕图像梯度方向边缘模型　出处：《华侨大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着移动互联网、云计算以及物联网等技术的快速发展,屏幕图像获得了越来越多的关注和应用。与自然图像不同,屏幕图像既包含由计算机直接生成或渲染的非连续色调内容,如文字、图表和图形等,又含有由摄像机采集得到的连续色调内容,如自然场景图片和视频片段等。屏幕图像在获取、存储、编码、传输和显示过程中会不可避免地引入各种类型的失真从而导致图像视觉效果降低。因此,如何准确评价屏幕图像质量是屏幕图像技术领域一个至关重要的问题。现有的大部分图像质量评价算法和图像质量评价数据库都是针对自然图像而设计的,由于屏幕图像与自然图像的不同统计特性,他们并不能完全适用于屏幕图像。因此,本篇论文根据人眼视觉特征和屏幕图像特点系统探索了屏幕图像质量评价方法,包括基于梯度方向的屏幕图像质量评价方法、基于边缘模型的屏幕图像质量评价方法、通用屏幕图像质量评价数据库。具体如下:1、在分析了屏幕图像与自然图像梯度幅值的显著差异和梯度方向与人类视觉系统相关性的基础上,设计了一种基于梯度方向的全参考型屏幕图像质量评价算法。该算法首先提出一种能够有效捕获屏幕图像失真特性的梯度方向计算方法,然后计算参考和失真屏幕图像的梯度方向相似度,进而融合梯度幅值相似度以获得最终的失真屏幕图像质量指标。实验结果表明,提出的方法能够有效准确地评价屏幕图像感知质量。2、考虑到人眼视觉对于图像边缘比较敏感且屏幕图像含有大量边缘,提出了一种基于参数化边缘模型的全参考型屏幕图像质量评价方法。该方法首先通过边缘模型分别从参考和失真屏幕图像中提取两种显著的边缘属性:边缘对比度和边缘宽度,然后分别计算参考和失真屏幕图像的边缘对比度相似度和边缘宽度相似度,最后使用基于边缘宽度的权重合并方法计算失真屏幕图像质量评价分数。实验结果证明,所提方法与人类视觉系统对屏幕图像的主观感知一致性较好。3、考虑到现有的最大屏幕图像质量评价数据库存在参考图像和失真类型不够全面的问题,建立了一个目前最大的主观屏幕图像质量评价数据库SCID。根据屏幕图像和其实际应用场景特点选取了40张参考屏幕图像,并使用9种常见的失真类型生成1800张失真屏幕图像。严格遵照国际标准采用双刺激损伤测量法获取失真屏幕图像的主观质量评分,经过分析与处理后得到每张失真屏幕图像的MOS值作为感知质量的真实值。基于所构建的SCID,我们深入评估了13种具有代表性的客观图像质量评价模型,并进行了详尽分析。该数据库将通过开放共享作为屏幕图像领域的一个重要数据库,对促进屏幕图像技术领域的研究工作具有重要意义。综上所述,本文主要根据屏幕图像特点和人眼视觉特性分析与探索屏幕图像质量评价方法,具有一定的创新性和挑战性。本文的研究成果在一定程度上为基于人类视觉系统的屏幕图像技术开拓了视野,具有重要的理论研究意义和实际应用价值。
[Abstract]:With the rapid development of mobile Internet, cloud computing and networking technologies, the screen image gained more and more attention. Different from the natural image, non continuous tone, the screen image contains not only directly generated by the computer or rendering such as text, graphics and graphics, but also contains the content of continuous tone collected by the camera the natural scene, such as pictures and video clips. The screen image acquisition, storage, encoding, transmission and display process will inevitably introduce various types of distortion resulting image visual effect is reduced. Therefore, how to accurately evaluate the screen image quality is one of the most important technology in the field of screen image and image evaluation algorithm. The quality of most of the image quality evaluation is based on the existing database of natural images and design, because the screen images and natural images with different statistical characteristics And they are not fully suitable for the screen image. Therefore, this thesis explores the image quality evaluation method based on human visual characteristics of the screen and the screen image features of the system, including the screen image quality evaluation method based on the gradient direction, quality evaluation method of edge model based on screen image, general screen image quality evaluation database as follows. 1, based on the analysis of the significant differences between the screen image and natural image gradient amplitude and gradient direction and the correlation of the human visual system, an algorithm of full reference image quality assessment screen based on gradient direction design. The algorithm first proposed can effectively capture a screen image distortion gradient direction calculation method. Then calculate the gradient direction similarity reference and the distortion of screen image, and then the fusion gradient amplitude similarity in order to get lost really the final screen The image quality index. The experimental results show that the proposed method can effectively and accurately evaluate the perceived quality of the screen image of.2, taking into account the human visual system for image edge sensitive and the screen image contains a large number of edges, presents a screen full reference image quality assessment based on parametric edge model method. This method firstly edge model respectively. From the reference and distortion from two significant attributes: the width of the screen image edge edge contrast and edge, and then calculate the width of edge contrast similarity and edge similarity reference and distortion of the screen image, finally use the distortion of image quality evaluation score screen to calculate the weight combination method based on the width of the edge. The experimental results show that the proposed method with the human visual system to screen image perceived consistent.3, taking into account the existing maximum screen image quality The evaluation database has the reference image and the distortion type is not comprehensive, set up a current evaluation of subjective image quality database SCID. the largest screen according to the screen image and its application scene features selected 40 a reference screen image, and use the 9 common types of distortion generated 1800 distortion screen image in strict accordance with international standards. Double stimulation damage measurement method to obtain the subjective quality score distortion of the screen image, after analysis and processing after each screen image distortion MOS value as the perceived quality of the true value. The construction of SCID based on our in-depth evaluation of the 13 kinds of objective image quality evaluation of representative models, and analyzes in detail the. The database will share an important database as the screen image field through the open, is to promote the research work of the technical field of the screen image Important. To sum up, this paper mainly based on the analysis and exploration of image quality evaluation method of screen screen image features and features of human vision, with innovative and challenging. The research results of this paper to a certain extent for the screen images of the human visual system to develop a vision based research has important theoretical significance and practical application value.

【学位授予单位】：华侨大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】