一种改进的无参考图像质量评估算法

摘要

无参考图像质量评估方法一直是图像质量研究所关注的重点。本文针对无参考图像质量评估问题，在当前被广泛接受的 DIIVINE 图像质量评估方法基础上进行了深入的研究，研究发现经典的 DIIVINE 方法中使用了过多的统计特征量，由于统计量无差别的使用影响了方法的预测性能。基于此，利用主成分分析（PCA）算法对该特征向量进行降维，得到“精简的特征向量”，用此向量可以达到更好的分类性能和评估性能。试验证明，在图像数据库上，该算法表现出较好的预测性能，其性能优于其他主流的无参考图像评估算法。%No reference (NR) image quality assessment method has been paid more attain in image quality research. For NR quality assessment problem, this paper focuses on the basis of DIIVINE which is a widely accepted image quality assess-ment. We firel that there are many useless statistical characteristic in DIIVINE, and the method's predictive performance has been affected as these characteristic introduced. Based on this, this paper utilizes principal component analysis ( PCA) method to reduce the dimension of the feature vector in DIIVINE, and then we use reduced feature vector to construct new prediction algorithm. It is validated the method on the image database, the method expresses a high predictive performance, which is superior to other famous no-reference image evaluation algorithms.