基于AdaBoost BP神经网络的无参考图像质量评价的研究

提出了一种利用相关梯度特征和AdaBoost 反向传播神经网络的无参考图像质量评价方法。首先利用高斯滤波器得到水平和竖直的方向导数，然后提取相关的梯度特征，其次计算其直方图方差特征，然后训练BP神经网络得到弱分类器并利用AdaBoost算法获得最终的强分类器，最后利用得到的强分类器预测图像质量分数。实验结果表明，方法评价的结果合理、鲁棒性强、实行性好，符合人类视觉特性，并且与主观评分有较好的一致性，取得了很好的评价效果。

Blind image quality assessment based on AdaBoost BP neural network

This paper presents a non reference image quality assessment method based on correlation gradient feature and AdaBoost back propagation (BP) neural network. In this method, firstly the Gaussian filter is used to obtain the horizontal and vertical direction derivative, then the relevant gradient feature is extracted, then calculate the histogram variance, and then the BP neural network is trained to obtain Liu Bin the weak classifier, Finally, strong classifier is obtained by AdaBoost algorithm. The image quality scores is predicted based on the strong classifier. The experiment results show that the results of this method are reasonable, stable, consistency with human visual characteristics, and subjective score has a good consistency. This method is simple and achieved good evaluation results.