无参考的特征点复杂度激光干扰图像评估  

No-reference laser-dazzling image quality assessment based on feature-point complexity

作　　者：钱方[1,2] 孙涛[1] 郭劲[1] 王挺峰[1] 

机构地区：[1]中国科学院长春光学精密机械与物理研究所激光与物质相互作用国家重点实验室,吉林长春130033 [2]中国科学院大学,北京100049

出　　处：《光学精密工程》2015年第4期1179-1186,共8页Optics and Precision Engineering

基　　金：激光与物质相互作用国家重点实验室科研基金资助项目(No.SKLLIM1203-01);吉林省重大科技攻关专项资助项目(No.20126015)

摘　　要：由于目标识别和检测效果取决于目标提取的准确性,本文结合主动成像与目标识别技术搭建了激光主动成像系统实验平台,研究了激光干扰对获取图像质量以及特征点提取效果的影响。提出了一种无参考的特征点复杂度图像评估算法。该算法在目标区域位置计算图像的特征点、纹理、梯度和对比度复杂度,然后综合4个因子得到归一化的评估结果。利用激光主动成像系统对设定目标进行了照明成像实验,同时采集了不同干扰功率和光斑位置的干扰图像。使用本文提出的特征点复杂度算法对标准数据库及实验获得的激光干扰图像进行了评估。实验结果表明:提出的评估算法能够客观地反映图像质量的变化情况,可对不同程度的激光干扰图像给出合理的评估结果,其评价结果更符合主观视觉感受,并且能够指导激光主动成像识别系统的防护与应用。The validity of target recognition and target detection depend on the accuracy of target extraction,this paper establishes an experiment platform of laser active imaging system combining laser active image technology and target recognition technology.The image quality and feature-point extraction were researched in the target area after laser jamming,and a new No-reference Featurepoint Complexity Metrics（NRFPCM）assessment algorithm was proposed.In the algorithm,the feature-point,texture,gradient and contrast complexity were computed in the target area,then the normalized NRFPCM was obtained via production of the four factors above.Finally,the luminance imaging experiment was performed for the target by utilizing the laser active imaging system and the disturbed images with different disturbing powers and different spot positions were obtained.Theproposed NRFPCM algorithm was used to evaluate the standard image database and the newly obtained laser-dazzling images.The results show that the proposed algorithm reflects the image quality objectively and gives a more reasonable evaluation results for different laser-dazzling images.Moreover,evaluation results are more suitable for the subjective visual feeling.The NRFPCM gives the guidance to the laser active imaging system in the defense and application.

关 键 词：图像处理 图像质量评价 激光干扰 特征点相似度 目标识别 

分 类 号：TN977[电子电信—信号与信息处理] TP391.4[电子电信—信息与通信工程]