基于局部保留投影和极端学习机的立体图像质量评价
首发时间:2015-12-01
摘要:为解决极端学习机(Extreme Learning Machine, ELM)由于初始权值和隐藏层节点偏移量的随机性引起的预测精度下降问题,本文提出了一种将遗传算法与ELM网络结构相结合的GA-ELM方法,对立体图像质量进行客观评价。ELM算法是单隐层前馈神经网络的泛化,本文通过GA优化ELM网络结构的初始权重与隐藏层节点偏移量,使优化后的ELM网络具有更好的分类识别效果。实验结果表明,使用不同激励函数的条件下,GA-ELM的图像质量等级正确分类率和所需的隐藏层节点个数均优于ELM。以sigmoid为激励函数,对256幅不同等级的立体图像测试样本进行测试,其正确等级分类率达到96.09%,优于现有算法。
关键词: 模式识别 客观评价 立体图像 局部保留投影 极端学习机
For information in English, please click here
Stereo image quality assessment based on LPP and ELM
Abstract:Currently, objective quality assessment of stereoscopic images is still in development, establishing an effective objective assessment in accord with subjective image quality assessment is still a problem. Since a three-dimensional image contains more information than a 2D image, effective feature extraction is greatly needed for stereoscopic images. Therefore, locality preserving projection method is adopted in this paper to reduce dimension and redundant information of images. At the same time, we select the extreme learning machine as classifier of the quality assessment model to classify images having been extracted features. The experimental results show that the objective assessment results of LPP-ELM algorithm can reach an accuracy of 93.2%, which means it can reflect subjective assessment of stereoscopic images accurately and effectively.
Keywords: pattern recognition objective assessment stereoscopic image locality preserving projection extreme learning machine
论文图表:
引用
No.4665869111665814****
同行评议
勘误表
基于局部保留投影和极端学习机的立体图像质量评价
评论
全部评论0/1000