基于图像质量和注意力的宫腔镜视频关键帧提取  

Key Frame Extraction of Hysteroscopy Videos Based on Image Quality and Attention

作　　者：苗强强 Miao Qiangqiang(College of Information Engineering,Wuhan University of Technology,Wuhan,Hubei 430070,China)

机构地区：[1]武汉理工大学信息工程学院,湖北武汉430070

出　　处：《激光与光电子学进展》2019年第6期122-128,共7页Laser & Optoelectronics Progress

基　　金：国家自然科学基金(61303028)

摘　　要：为解决传统方法中由注意力曲线等长分段导致的精确率偏低的问题,提出了基于图像质量曲线提取关键视频段的方法,从各段内选取注意力值最高的帧作为关键帧。针对本地数据库进行视频关键帧提取,精确率为52.94%,F值为62.77%,分别比Muhammad的方法提高了5.23%和2.65%。In order to solve the problem of low accuracy rate caused by the equal length segmentation of the attention curve in the traditional methods, a scheme is proposed, in which the key video segments are extracted based on the image quality curve. The frame with the highest attention value is selected from each segment as the key frame. On the local database, the precision and F-measure are 52.94% and 62.77%, 5.23% and 2.65% higher than those by the Muhammad method, respectively.

关 键 词：图像处理 宫腔镜视频摘要 注意力模型 图像质量评估 图像锐利度 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]