The discrepancy between an image and its “reblurred” version indicates the extent of blur in the image. This paper presents a novel no-reference image sharpness evaluator leveraging the discrepancy measures of structural degradation in both the spatial and wavelet domains. Specifically, local structural degradation of an input image is characterized by the discrepancy measures of orientation selectivity-based visual patterns and log-Gabor filter responses between the image and its corresponding reblurred version respectively. Considering the influence of viewing distance on image quality, the global sharpness discrepancy is measured through inter-resolution self-similarities. Finally, the computed discrepancies are utilized as sharpness-aware features and then a support vector regressor is employed to map the feature vectors into quality scores. The performance of the proposed method is evaluated on six public image quality databases, including two real blurred image databases. Experimental results demonstrate that our proposed method achieves state-of-the-art performances across all these databases.

图像与其“重新模糊”版本之间的差异表明图像中的模糊程度。本文提出了一种新颖的无参考图像清晰度评估器，该评估器利用了空间域和小波域中结构退化的差异度量。具体地，输入图像的局部结构退化的特征在于分别基于图像和其对应的模糊版本的基于方向选择性的视觉模式和log-Gabor滤波器响应的差异度量。考虑到观看距离对图像质量的影响，通过分辨率间的自相似性来测量全局清晰度差异。最后，将计算出的差异用作锐度感知特征，然后使用支持向量回归器将特征向量映射为质量得分。在六个公共图像质量数据库（包括两个真实的模糊图像数据库）上评估了该方法的性能。实验结果表明，我们提出的方法可在所有这些数据库中实现最先进的性能。