面向应用需求的可见光与红外彩色融合图像综合质量评价

The visible and infrared color fusion technology helps observers to identify targets and understand scene contents, and it shows extensive application prospects in many fields, such as military reconnaissance and vehicle navigation. Different fusion algorithms meet specific application requirements at different levels. How to evaluate fusion image quality objectively based on application requirements without reference becomes one of pressing problems. This project will establish a comprehensive quality assessment system for color fusion images based on visual tasks: (1) Based on psychophysical experiments, color fusion image database and image quality subjective assessment database will be established. The assessment metrics include comprehensive quality based on application requirements (comprehensive quality based on scene understanding and comprehensive quality based on target detectability) and four single quality attributes (color harmony, sharpness of the scene, sharpness of the target and perceptual contrast between the target and background). The prediction models of the comprehensive quality will be estimated based on these quality attributes by analyzing the subjective assessment data; (2) Considering human visual characteristics, the no-reference objective assessment models of these quality attributes will be established, whose predictions will be more closely matched to human perceptual evaluations. (3) The no-reference comprehensive quality assessment for color fusion images based on application requirements will be obtained by combining the quality attribute models. The validity of the comprehensive quality assessment models will be test by comparing to the subjective assessment data. This research will solve the difficult problem of color fusion image comprehensive quality objective assessment and promote the development and application of color fusion technology.

可见光与红外图像彩色融合技术有助于观察者探测目标并理解场景内容，在军事侦察、车辆导航等众多领域展现出广阔的应用前景。不同融合算法满足特定应用需求的能力不同，如何针对应用需求客观无参考地评价融合图像质量成为目前亟待解决的难题之一。项目拟建立一套面向应用需求的融合图像综合质量评价体系：⑴拟通过主观评价实验，建立彩色融合图像库及其主观评分数据库，包括面向应用需求的综合质量(基于场景理解和基于目标探测的综合质量)和单一质量(颜色协调性、目标清晰度、场景清晰度和目标背景感知对比度)。分析实验数据，利用四项单一质量构建综合质量预测模型；⑵结合人眼视觉特性，建立各单一质量无参考客观评价模型，达到与人眼主观感受更一致的评价效果；⑶结合单一质量客观评价模型，构造面向应用需求的综合质量客观评价模型，并用主观评分数据验证其有效性。有望解决彩色融合图像综合质量无参考客观评价的难题，促进彩色融合技术的发展与应用。

可见光与红外彩色融合系统已经开始装备应用，进入实用化阶段。然而，目前国内外尚没有公认的彩色融合图像质量的评价理论和方法，直接影响到对彩色融合成像系统的评价，成为技术研究和系统应用迫切需要解决的焦点问题之一。项目建立了一套面向应用需求的融合图像综合质量评价体系，具体包括：(1)通过主观评价实验，建立彩色融合图像库及主观评分数据库，评价指标包括面向应用需求的综合质量(基于场景理解和基于目标探测的综合质量)和单一质量(颜色协调性、目标清晰度、场景清晰度和目标背景感知对比度)。(2)考虑相邻图像的质量评价对当前图像评价结果的影响，采用子空间辨识方法建立融合图像综合质量高阶预测模型。将单一质量分数作为输入项，综合质量评价结果作为输出项，通过输入、输出Hankel矩阵投影的行“子空间”和列“子空间”来获取模型参数。通过对实验结果的分析处理，验证了该方法的有效性。(3)建立了基于联合特征的彩色融合图像颜色协调性评价模型。采用记忆色理论计算局部图像的颜色协调性分数解决了参考图像缺失的问题，通过隐含狄利克雷分布获得融合图像的全局特征，从局部图像的局部特征和全局特征两个方面对基础网络进行训练，通过所有局部图像的预测分数对彩色融合图像颜色协调性进行客观评价。实验结果表明，该模型的计算结果与人眼主观感受具有较高的一致性。(4)综合考虑人眼的对比度敏感特性以及亮度掩盖特性，从亮度对比度和基于归一化色差模型的颜色对比度两方面来衡量目标与背景的感知对比度。模型计算结果可以较好地反映人眼视觉感受。(5)前期研究已建立了场景清晰度和目标清晰度评价模型，结合本项目建立的单一质量客观评价模型，利用基于子空间辨识的图像综合质量高阶预测模型，实现面向应用需求的图像综合质量无参考客观评价，并达到与人眼主观感受较一致的评价效果。解决彩色融合图像综合质量无参考客观评价的难题，促进彩色融合技术的发展与应用。