In order to evaluate the image quality objectively,a no-reference image quality evaluation model,which combined color space statistical feature and weighted local binary model(LBP)texture features,was established in this paper.Firstly,the mean subtracted contrast normalized(MSCN)co-efficients were obtained by applying MSCN to the luminance of the distorted image.Then,the statis-tical parameters and weighted LBP histogram features were extracted from the MSCN coefficients. The statistical parameters were obtained from the generalized Gaussian model and the weight was the magnitude of the MSCN coefficients.In addition,the natural scene statistics(NSS)feature of red-green and blue-yellow components in Lαβcolor space was used to enhance the description based on col-or distortion,and the statistical feature parameters were obtained by using asymmetric generalized Gaussian model.Finally,SVR was used to establish the regression model that mapping image quality evaluation index to subjective quality score.The experimental results on LIVE,CSIQ,TID2013 and MLIVE databases showed that the weighted average Spearman rank correlation coefficient(SROCC) was 0.776 and the Pearson linear correlation coefficient(PLCC)was 0.821,which was superior to other methods.In addition,the extraction of 512× 512 size image feature spent only 0.19 s.The method proposed in this paper is consistent with the subjective perception of human eyes and has the advantage of low complexity.%为了客观评价图像质量,本文提出联合颜色空间统计特征和权重局部二值模式(Local Binary Pattern,LBP)纹理特征的无参考图像质量评价模型.首先,对失真图像进行亮度去均值对比度归一化(Mean Subtracted Contrast Normal-ized,MSCN)操作得到MSCN系数;然后,对MSCN系数提取其统计参数特征和权重LBP直方图特征,其中统计参数由广义高斯模型获得,权重为MSCN系数的幅度.另外,还采用了 Lαβ颜色空间下红绿和蓝黄分量的自然场景统计(Nat-ural Scence Statistics,NSS)特征来增强基于颜色失真的描述,并运用非对称广义高斯模型获得统计参数特征.最后,运用SVR建立图像质量评价指标到主观质量得分的回归模型.在LIVE,CSIQ,TID2013和MLIVE数据库上的实验结果表明:4个数据库加权平均Spearman秩相关系数为0.776,Pearson线性相关系数为0.821,均优于其他方法;图像大小为512×512时特征提取只需0.19 s.本文提出的方法与人眼主观感知具有良好的一致性,并具有复杂度低等优点.
展开▼
