根据人类视觉系统(Human Visual System, HVS)中双目视觉信息处理的过程, 结合一系列图像特征, 提出一种基于双目信息融合的立体图像质量评价模型。该模型通过复小波变换模拟HVS对立体图像的融合过程。提取结构活跃度(Structural Activity, SA)以及相位一致性(Phase Congruency, PC)作为图像特征。最后通过度量融合图像特征的改变程度获得立体图像客观质量。采用本文所提出的客观评价模型对立体图像数据库进行评价, 其线性相关系数值在0.92以上, 均方根误差值接近6, 异常值比率值为0%。实验结果表明, 该模型符合人眼视觉特性, 能够很好地预测立体图像质量。

According to the procedure of binocular information processing in human visual system (HVS), combining with a series of image features, a new stereo image quality metric based on binocular fusion is proposed. In this model, firstly, complex wavelet transform is utilized to mimic HVS mechanism to fuse stereo image pair. Secondly, two important features, structural activity and phase congruency, are extracted as fusion image features. Finally, the difference of fusion image feature is calculated as stereo image quality. By applying the proposed model to stereo image test database, it is experimentally demonstrate that correlation coefficients are all above 0.92, the RMSE approaching to 6 and the OR staying at 0%. It is indicated that the proposed model is consisted with human perception and stereo image quality can be evaluated accurately.