文献类型：期刊文献

中文题名：一种无参考监控视频图像清晰度评价方法

英文题名：A No-Reference Surveillance Video Image Sharpness Assessment

作者：邱铭杰[1];常青[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2014

卷号：40

期号：4

起止页码：465

中文期刊名：华东理工大学学报（自然科学版）

外文期刊名：Journal of East China University of Science and Technology

收录：CSTPCD;;Scopus;北大核心:【北大核心2011】；CSCD:【CSCD2013_2014】；

语种：中文

中文关键词：图像清晰度;结构相似度(SSIM);无参考;监控视频

外文关键词：image sharpness; SSIM; no-reference; surveillance video

摘要：监控视频调焦时容易出现镜头离焦,导致输出图像模糊。为了降低人工检测成本,提出了一种基于结构相似度的无参考图像清晰度评价算法,用于评价监控安防设备视频图像的清晰度。在无参考的情况下,首先利用Sobel算子和高斯滤波器对待评价图像构造参考图,在此基础上计算分块后的图像结构相似度SSIM,为了更好地与人眼习惯吻合,利用权值加成各子块SSIM值,最后根据计算值评价图像的清晰度。实验结果表明:本文方法对图像模糊敏感度高,检测准确性优于传统检测方法,可以应用于实时的监控视频图像诊断。
Blurred images often appear in the course of surveillance video lens focusing. In order to reduce the cost of artificial detection, a no-reference new method based on SSIM is proposed to assess the sharpness of surveillance video image, In the case of no reference, both Sobel operator and Gaussian lowpass filter are firstly utilized to construct the reference image, and then, the SSIM of partitioned image is computed. Moreover, the quality of the image is judged via the weighted value of sub block SSIMs. The experiment results show that the proposed algorithm in this work is highly sensitive to burred image and can achieve higher accuracy than traditional methods. Hence, it can be applied to the real-time surveillance video image diagnosis.