基于仿视觉细胞模型的立体图像质量评价方法

Stereoscopic Image-Quality-Assessment Method Based on Visual Cell Model

在观看立体图像时,人类视觉系统(HVS)以人眼视网膜细胞为媒介接收、传输和理解双目信息,基于此提出了一种仿视觉细胞模型的立体图像质量评价(SIQA)方法。以视网膜细胞特性为基础,模仿HVS对简单细胞和复杂细胞进行建模;双目信息经简单-复杂细胞模型处理得到双目融合视点图(BFVM)和双目细胞差异图(BCDM),采用多尺度结构相似度(MSSIM)算法分别对原始、失真立体图像的BFVM和BCDM进行双目融合评价(BFQA)和双目立体感评价(BSPA);对BFQA和BSPA的评价值进行加权融合得到最终评价值。实验结果表明,该方法的Pearson线性相关系数在0.94以上,Spearman秩相关系数在0.93以上,该模型符合人眼视觉特性,能够较好地预测立体图像质量。

Retinal cells of human visual system（HVS） are used as a medium to receive, transmit and understand binocular information when stereo images are viewed. A new stereo image-quality-assessment（SIQA） algorithm aiming to imitate retinal cells is proposed. HVS is mimicked by modeling simple cells and complex cells on the basis of the retinal cell characteristics. Binocular fusion view map（BFVM） and binocular cell-based disparity map（BCDM）are obtained via the binocular information processing of the simple-complex cell models. Based on the multiscale structural similarity（MSSIM） method, the binocular fusion quality assessment（BFQA） and binocular stereo perception assessment（BSPA） are respectively implemented to the original or distorted stereo images. The results of BFQA and BSPA are combined to describe the final quality of stereo images. The experimental results demonstrate that the overall Pearson linear correlation indicator reaches 0.94, Spearman rank order correlation coefficient reaches0.93, which indicate that the proposed model is well consistent with human perception, and also well predict the stero image quality.