基于SEM的彩色融合图像质量主观评价方法

Subjective assessment of color fused image quality based on SEM

作为客观评价的研究基础,人眼主观评价是比较不同融合算法图像优劣的最直接有效的方法。通过开展主观评价实验,基于单因变量模型分析清晰度、颜色协调性等影响因素对彩色融合图像质量的影响程度,进而预测视觉任务下的融合图像总体质量,是目前流行的彩色融合图像质量主观评价研究方法。针对基于单因变量模型的研究方法易割裂“探测”与“感知”在探测识别过程中的联系的问题,本文以村庄环境下的探测识别视觉任务为例,选取“感知”和“探测”2个指标共同作为彩色融合图像质量的表现指标,组织15人对由8个场景、8种典型融合算法生成的42幅彩色融合图像的质量进行了主观评价。采用双因变量的结构方程模型(Structural Equation Model,SEM)分析了筛选后的评价结果,建立了彩色融合图像质量的预测模型并验证了模型的正确性。结果表明,当同时表现“感知”和“探测”2种性能时,村庄环境、探测识别视觉任务下,彩色融合图像的质量可由“目标背景差异”、“场景清晰度”和“颜色协调性”3个因素预测,且“感知”和“探测”2种性能是均衡的。

As the research foundation of the objective assessment,the subjective assessment of human eyes is the most direct and effective method to compare the image quality of different fusion algorithms.It is a popular research method for subjective assessment of color fused image quality at present to analyze impacts of factors as the sharpness and the color harmony on the color fused image quality based on the single dependent variable model by carrying out subjective assessment experiments,and then to predict the overall quality of fused images under vision tasks.Aiming at the problem that the method based on the single dependent variable model tends to cut apart the relationship between"detection"and"perception"in the process of detection and recognition,the vision task of detection and recognition in the village is taken as an example in this paper,"perception"and"detection"are selected as the externalization indicators of the color fused image quality.Fifteen people are organized to subjectively assess the quality of forty two color fused images generated by eight typical fusion algorithms in eight scenes.The assessment results after screening are analyzed by the structural equation model(SEM)of dual dependent variables.The prediction model of color fused image quality is established and the correctness of the model is verified.The results show that the color fused image quality under the detection and recognition vision task in the village environment can be predicted by three factors of"difference between targets and the background",and the two"scene sharpness"and"color harmony"are balanced when"perception"and"detection"are performed simultaneously.