弹载侦察图像质量评价方法研究

Research on On-board Reconnaissance Images Quality Assessment

针对弹载侦察图像由于受成像平台和条件限制存在多类型混杂失真,为了对图像后续处理和成像系统性能优化提供重要量化依据和指标参考,开展此类图像质量评价研究。分析了弹载侦察图像的特点,针对成像平台的多自由度、多姿态变化情况,提取与相机抖动、结构变化、颜色丢失相关的3类11种图像特征,利用高分辨率测绘图像作为原始图像集,分块提取特征进行多元高斯模型(MVG)拟合获得基准分布特性,对弹载图像也进行分块处理获得对应MVG分布特性,比较其与基准MVG特性马氏距离作为子块质量得分,计算各子块的得分均值作为整幅图像最终得分。以实际弹载成像平台获取的图像进行了实验验证。实验结果表明:与盲图像质量测度算法、基于离散余弦变换的图像完整性测度改进算法、基于失真度识别的图像验证与完整性评价算法、质量感知聚类算法、盲图像质量空域测度算法、自然图像质量测度算法、整体与局部的自然图像质量测度算法7种其他经典方法相比,该算法具有更好的准确度和主客观一致性。

Many types of hybrid distortion of on-board reconnaissance images happen due to the imaging platform and constraints. The image quality assessment is carried out in order to provide an important quantitative basis and reference for the performance optimization of the image processing and imaging sys- tems. The characteristics of on-board reconnaissance image are analyzed. In consideration of mutli-de- gree-of-freedom and multi-attitude change of imaging platform, 11 kinds of image features related to cam- era shake, structure change and color loss are extracted. The sub-block image features are extracted for multivariate Gaussian （MVG） fitting to obtion benchmark distribution characteristics by using high reso- lution mapping images as original image set, and the on-board reconnaissance images are also processed to obtain corresponding MVG distribution characteristics. The benchmark distribution characteristics are compared to the MVG distribution characteristics, and Mahalanobis distance is calculated as a score of block quality. The mean score of each sub-block is taken as final score of whole image. The images ob- tained by the actual on-board imaging platform are verified. The result shows that the proposed method has higher assessment accuracy.