集成学习算法在图像质量评估中的应用

Application of ensemble learning algorithm in image quality assessment

本文基于方法融合感知的思想,探究了集成学习算法在图像质量评估中的应用,对利用支持向量机及神经网络搭建集成学习框架进行图像质量评估的性能表现进行了分析。为保证集成学习后评估算法的通用性,使用保真度、结构相似性、颜色质量与数据驱动学习等多种估计器指标进行集成,使用LIVE、多重失真LIVE及TID 2013 3种数据库进行验证。实验结果表明,集成学习方法通常可以提高图像质量评估能力,基于深度学习网络的评估能力增强优于基于支持向量机的增强,当存在两种以上附加方法进行融合感知时,二者较原有最佳性能方法均可取得较为有效地提升。

Based on the idea of method fusion perception,this paper explores the application of ensemble learning algorithm in image quality assessment,and analyzes the performance of an ensemble learning framework for image quality assessment combining support vector machine and neural network.In order to ensure the universality of the post evaluation algorithm of ensemble learning,this paper uses a variety of estimator indicators such as fidelity,structural similarity,color quality and data-driven learning.In the experiment,the LIVE,the multiply distorted LIVE,and the TID 2013 databases are used for verification.The experimental results show that ensemble learning methods can generally improve the image quality evaluation ability.The enhancement of evaluation ability based on deep learning network is better than that based on support vector machine.When there are more than two additional methods for fusion perception,both of them can achieve better effect than the original best performance method.