中文摘要：

针对现有图像梯度特征质量评价方法的不足,充分考虑人类视觉系统（HVS）特性,提出了一种基于HVS多通道梯度和低阶矩联合（MGLJA）方法。MGLJA基于log-Gabor滤波器提取图像的多通道视觉特征并予以阈值稀疏化处理;然后以梯度相似度算法和低阶矩失真度算法分别评价各通道视觉特征的能量特征和结构特征;在评价融合阶段,MGLJA首先依据视觉中央凹空域加权局部评价,其次依据对比度函数频域加权各通道评价;最后,通过一步快速回归算法自适应地融合梯度评价和低阶矩评价得到待测图像的质量评价。不同类型和水平失真图像的实验结果表明,MGLJA方法的斯皮尔曼原始排序相关系数、回归曲线的相关系数及均方根误差等评价指标均优于PSNR类传统方法,与特征结构相似度（FSIM）及视觉显著（VSI）等新方法相比整体性能上也有较大优势,且评价实时性也明显优于FSIM方法。

英文摘要：

Aiming at the defect of image quality assessment methods with image gradient characteristic,sufficiently considering the property of human visual system（ HVS）,a novel method based on HVS multi-channel gradient and low-order-moment joint assessment（ MGLJA） is proposed. This method extracts the visual multi-channel feature of the images using log-Gabor filter and sparsely processes the extracted feature with visual threshold. Local visual gradient similarity and low order moment distortion algorithms are used respectively to assess the energy feature and structural distribution feature of every channel. In the stage of assessment pooling,firstly,the local assessment is averaged with the weight of visual fovea; then,the assessment of the channels is conducted with the weight of the contrast sensitive function; finally,the overall quality assessment of the distorted image is achieved by adaptively integrating the gradient similarity and structural distribution distortion assessment with one step quick regress algorithm. The experiment results of the images with the distortion of different types and levels show that the proposed MGLJA method is superior to the traditional PSNR methods in terms of spearman rank-order correlation coefficient（ SROCC）,correlation coefficient（ CC） of the regression curve and root mean square error（ RMSE）; compared with the new methods,such as feature similarity（ FSIM）,visual saliency induced（ VSI） and other relevant state-of the-art image quality assessment methods,the MGLJA method also has advantage in overall performance. Furthermore,the real-time performance of the MGLJA method is obviously superior to that of the FSIM method.