高级检索

结合图像二次模糊范围和奇异值分解的无参考模糊图像质量评价

No-reference Quality Assessment for Blur Image Combined with Re-blur Range and Singular Value Decomposition

  • 摘要: 针对图像的模糊失真问题,提出一种结合图像二次模糊范围和奇异值分解的无参考图像质量评价方法.首先根据图像二次模糊范围的差异性构建参考图像;然后对失真图像与参考图像进行奇异值分解,利用奇异值向量矩阵的相似度构建失真特性向量;再结合Log-Gabor滤波器组和高斯差分模型进行视觉显著度检测;最终以显著度加权的失真特性向量预测模糊图像的质量得分.大量实验结果表明,与同类算法相比,文中方法能够准确地评价模糊图像质量,与主观评价具有较高的一致性;在LIVE2图像库上,评价指标斯皮尔曼等级相关系数达到0.968 7,均方根误差为4.858 9;该方法无需对数据进行训练,具有较强的实用性和推广性.

     

    Abstract: In this paper, a new no-reference image quality assessment(IQA) algorithm is proposed for blur images based on re-blur range and singular value decomposition. Firstly, a reference image is constructed based on the difference of re-blur range. Secondly, singular value decomposition is utilized on reference image and test image, and distortion feature vectors are extracted by computing the comparability of singular value matrixes. Thirdly, visual saliency is detected by Log-Gabor filters and the difference of Gaussian model. Finally, the image quality is assessed from distortion feature vectors weight by visual saliency. Extensive experiments conducted on publicly IQA databases demonstrate that this method has higher correlation with human judgment and obtains a better evaluation index compared to other methods. The performance indices of Spearman rank correlation coefficient and root mean square errors on the LIVE2 database are 0.968 7 and 4.858 9, respectively. It doesn’t need training to assess image quality and has wide value for application and popularization.

     

/

返回文章
返回