基于广义回归神经网络无参考模糊图像质量评价

摘要

提出一种基于广义回归神经网络的无参考模糊图像质量评价方法.该方法首先通过相位一致变换生成待评测图像的相位一致图像,然后利用灰度共生矩阵计算相位一致图像的信息熵、能量、对比度、相关性和同质性5个特征,最后利用广义回归神经网络模型训练学习,预测得到无参考模糊图像质量得分.在3个公开数据库模糊图像上的实验结果表明,新方法预测得分与主观得分有较好的一致性,更加符合人类视觉特性.%It is proposed a no-reference blur image quality assessment method that adopts general regression neural network(GRNN). Firstly,the method gets the phase congruency image by phase congruency transformation. Then it calculates the five characters of the image, such as the information entropy, the energy, the contrast, the correlation and the homogeneity by Gray level co-occurrence matrix. Trained by GRNN, it predicts the quality of the No-reference Blurred Image. Experimental results show that the prediction of the new method is consistent with the subjective ones well. The assessing result fits the human visual judgment better.