自动目标识别中的图像序列质量评价方法

图像质量评价是自动目标识别(ATR)性能评估中的重要组成部分.传统的评价指标如目标信噪比(SNR)等针对的皆是单幅图像,而对于图像序列质量评价的研究尚属空白.针对该问题,该文首次提出了"帧间目标变化程度"的概念,利用其定量描述图像序列的质量.该指标融合了3部分信息:图像序列中帧间目标区纹理变化的信息,帧间目标大小变化的信息以及帧间目标位置不规律的信息.为验证所提指标的有效性,设计了用于分析"帧间目标变化程度"和ATR算法实际效果关系的实验,实验用样本为真实的目标图像序列.实验结果表明,"帧间目标变化程度"与ATR算法性能具有很强的相关性,基本呈现单调关系,是一种有效的图像序列质量评价标准.

Quality Estimation of Image Sequence for Automatic Target Recognition

Image quality estimation is an important part of performance evaluation for Automatic Target Recognition (ATR). Traditional image metrics are focused on single image, and there is no valid method to estimate the quality of image sequence. According to the above problem, this paper proposes the concept “Inter-Frame Change Degree of Target (IFCDT)” which is used for quantitatively describing image sequence quality for the first time. There are three key elements in the formula of the proposed metric, which embody the information of inter-frame change of target image texture, inter-frame change of target size and inter-frame change of target position respectively. To validate this image sequence metric, this paper designs an experiment for analyzing the relationship of IFCDT and ATR algorithm performance, while the samples of this experiment are actual target image sequences. The experiment result shows that there is a good monotone relationship between the image sequence metric and algorithm actual performance. Therefore, it can be concluded that the metric is a valid criterion to evaluate the quality of target image sequence.