朱子岩,
林睿,
林臻,
廖彦剑,
重庆大学生物工程学院 ??重庆 ??400044
基金项目:科技部国家重点研发计划(2016YFC0107113)，重庆市重点产业共性关键技术创新专项(CSTC2015ZDCY-ZTZXX0002)
详细信息
作者简介:罗洪艳：女，1976年生，博士，副教授，研究方向为医学图像处理
朱子岩：男，1993年生，硕士生，研究方向为图像质量评价、数字全息成像
林睿：女，1995年生，硕士生，研究方向为图像质量评价、数字全息成像
林臻：女，1995年生，硕士生，研究方向为图像质量评价、数字全息成像
廖彦剑：男，1976年生，博士，副教授，研究方向为医疗仪器及医学图像处理
通讯作者:廖彦剑 azurelyj@163.com
中图分类号:TP391计量
文章访问数:778
HTML全文浏览量:288
PDF下载量:52
被引次数:0
出版历程
收稿日期:2018-02-28
修回日期:2018-08-13
网络出版日期:2018-08-21
刊出日期:2019-01-01
Improved No-reference Noisy Image Quality Assessment Based on Masking Effect and Gradient Information
Hongyan LUO,Ziyan ZHU,
Rui LIN,
Zhen LIN,
Yanjian LIAO,
Institute of Bioengineering, Chongqing University, Chongqing 400044, China
Funds:The National Key R & D Program of Ministry of Science and Technology (2016YFC0107113), The Generality Critical Technology Innovation Special Items of Key Industry in Chongqing (CSTC2015ZDCY-ZTZXX0002)
摘要
摘要:针对目前大多数噪声图像质量评价算法借助域变换或机器学习所带来的运算量大、训练过程繁复等弊端，以及依赖人工设置固定阈值存在普适性不佳的问题，该文改进了一种基于掩盖效应的空域噪声图像质量评价算法。首先依据Hosaka原理提出层递进的分块规则，将图像分成与其内容频率分布高低相符的不同尺寸的子块并赋予相应的掩盖权值；然后通过提取像素点梯度信息，经两步检噪实现子块噪点甄别；再使用掩盖权值对子块噪声污染指标加权得到初步质量评价结果；最终修正和归一化后为整图质量评价结果——改进的无参考峰值信噪比(MNRPSNR)。应用该算法在LIVE和TID2008图像质量评价数据库上对多种噪声类型图像进行实验，结果显示其较目前主流评价算法保有很强竞争力，对传统算法改进效果显著，与人眼主观感受一致性高，普适于多种噪声类型。
关键词:无参考图像质量评价/
掩盖效应/
噪声检测/
梯度信息
Abstract:Heavy computational burden, or complex training procedure and poor universality caused by the manual setting of the fixed thresholds are the main issues associated with most of the noise image quality evaluation algorithms using domain transformation or machine learning. As an attempt for solution, an improved spatial noisy image quality evaluation algorithm based on the masking effect is presented. Firstly, according to the layer-layer progressive rule based on Hosaka principle, an image is divided into sub-blocks with different sizes that match the frequency distribution of its content, and a masking weight is assigned to each sub-block correspondingly. Then the noise in the image is detected through the pixel gradient information extraction, via a two-step strategy. Following that, the preliminary evaluation value is obtained by using the masking weights to weight the noise pollution index of all the sub-blocks. Finally, the correction and normalization are carried out to generate the whole image quality evaluation parameter——i.e. Modified No-Reference Peak Signal to Noise Ratio (MNRPSNR). Such an algorithm is tested on LIVE and TID2008 image quality assessment database, covering a variety of noise types. The results indicate that compared with the current mainstream evaluation algorithms, it has strong competitiveness, and also has the significant effects in improving the traditional algorithm. Moreover, the high degree of consistency to the human subjective feelings and the applicability to multiple noise types are well demonstrated.
Key words:No-reference image quality assessment/
Masking effect/
Noise detection/
Gradient information
PDF全文下载地址:
https://jeit.ac.cn/article/exportPdf?id=7b4b3633-88d4-4cdc-bd56-86922a62ee09
