基于视觉显著失真度的图像质量自适应评价方法

针对结构相似(SSIM)图像质量评价算法没有考虑人眼视觉多通道性和对图像高失真评价的不稳定性，提出一种基于视觉显著失真度的图像质量自适应融合(VSAP)评价方法。该方法首先采用log-Gabor滤波提取图像的高频、中频及低频3层视觉特征，基于log-Gabor变换尺度和方向权重系数计算特征值的相似度；然后基于视觉阈值多分辨性迭加计算出特征值的失真度；最后，根据视觉失真度自适应融合相似度评价与失真度评价获得图像质量的最终客观评价。实验结果表明，VSAP方法不但对图像不同类型失真的客观评价与主观感知具有更高的相关性，而且3个主要指标斯皮尔曼等级相关系数(SROCC)、曲线拟合相关系数(CC)和均方根误差(RMSE)对图像不同水平失真的整体评价性能更稳定，明显优于其它评价方法。

Image Quality Self-adaptive Assessment Based on Visual Salience Distortion

The Structural SIMilarity (SSIM) algorithm of image quality assessment does not take into account the characteristics of multi-channel resolutions of human vision, it is also not consistent with subjective human evaluation for high level distortions. A Visual Salience Adaptive Pooling (VSAP) strategy of image quality assessment is proposed based on visual multi-scale and multi-orientation of log-Gabor transformation. Firstly, the visual characteristics of image on the high, medium, and low frequency are extracted by the log-Gabor transformation. Then the visual similarity scores based on visual scales and visual orientations of log-Gabor are calculated, accordingly, the visual distortion levels of image are calculated iteratively with the visual multi- resolution threshold. Finally, a strategy of image quality assessment is proposed with adaptive pooling similarity scores to distortion scores. The experimental results show that objective assessments of VSAP for different types of distortion hold higher correlation with subjective assessment. More importantly, the overall assessment performance of the Spearman Rank-Order Correlation Coefficient (SROCC), Correlation Coefficient (CC) and Root Mean Square Error (RMSE) for different levels of distortion is more consistent with subjective scores and superior to other methods.