灰度夜视融合图像质量结构相似度评价方法

Structural Similarity Evaluation Method to Night Vision Fusion Image Quality

针对灰度夜视融合图像质量评价问题,提出利用构造的参考图像与融合图像进行结构相似度比较的客观评价方法.分别利用均值、标准差、熵、平均梯度及空间频率等全像素统计指标作为构建参考图像的构造因子,利用权重协调输入源图像在参考图像中的比例关系,获得了融合图像质量加权结构相似度评价模型.考虑到结构相似度评价模型中的结构函数对于严重模糊或失真的图像不能很好地反映图像的结构变化,进一步提出了梯度结构相似度评价模型和平均联合熵结构相似度评价模型.利用主观评价实验比较提出的三种基于结构相似度的融合图像质量评价模型发现,平均联合熵结构相似度评价模型有更好的稳定性且更接近人眼主观感受.

A new objective assessment method was proposed by means of the comparison between the structural reference image and fusion image based on structural similarity evaluating model to solve the question of night vision fusion image quality evaluation.Firstly,the reference image in weighting structural similarity evaluating model was made by the structural factor of the image element statistical indexes including mean,standard deviation,entropy,averaging gradient and spatial frequency.Meanwhile,the weight factor was used to adjust the proportional relation of input source image in reference image.Furthermore,gradient structural similarity and average joint entropy structural similarity model was adapted for assessing images of different distortion and blurring types because the structural function in structural similarity model was unavailable to express the structure change under the worsening distortion condition.Subjective experiment result showed average joint entropy structural similarity model was agree with human visual features,which was more consistent with the subjective assessment result than traditional methods.