基于极限学习机的全参考立体图像质量评价

发布时间：2018-06-19 06:47

本文选题：立体图像 + 质量评价　；参考：《计算机辅助设计与图形学学报》2017年05期

【摘要】：立体图像质量评价是评价立体视频系统性能的有效途径,而模拟人类大脑神经网络进行特征提取是立体图像质量评价的关键.为此,提出一种基于极限学习机的全参考立体图像质量评价方法,包括3个阶段:1)对原始和失真立体图像分别进行特征描述,以左图像,右图像和独眼图作为输入信息,采用包含3个隐层的极限学习机将图像信息映射到特征空间,从而得到原始和失真立体图像的特征描述;2)对原始和失真立体图像的特征描述进行相似性度量,从而得到原始和失真立体图像的质量特征;3)采用极限学习机建立得到的12维质量特征与主观评价值的回归模型,并将训练得到的回归模型用于测试阶段,预测得到相应的客观质量评价值.实验结果表明,文中方法在对称和非对称立体图像数据库都取得了较好的性能,与人类的主观感知保持良好的一致性.
[Abstract]:Stereo image quality evaluation is an effective way to evaluate the performance of stereo video system, and feature extraction based on simulated human neural network is the key of stereo image quality evaluation. In this paper, a new method for evaluating the quality of full reference stereo images based on extreme learning machine is proposed, which includes three stages: 1) describing the original and distorted stereo images respectively, taking the left image, the right image and the one-eye image as the input information. The image information is mapped to the feature space by an extreme learning machine which contains three hidden layers, and the feature description of the original and distorted stereo images is obtained. (2) the similarity between the original and the distorted stereo images is measured. Thus the quality characteristics of the original and distorted stereo images are obtained. The regression model of 12 dimensional quality features and subjective evaluation values established by the extreme learning machine is used in the test stage. The corresponding objective quality evaluation value is obtained by prediction. The experimental results show that the proposed method achieves good performance in both symmetric and asymmetric stereo image databases and is in good agreement with human subjective perception.
【作者单位】：宁波大学信息科学与工程学院;
【基金】：国家自然科学基金(61271021)
【分类号】：TP391.41

【相似文献】