摘要
在Yong等人提出的人类感光机制理论的基础上,提出一种基于Young-Helmholtz三色代数模型的RGB图像压缩方法.首先构建Young-Helmholtz三色代数框架下的GHA网络,抽取RGB图像的各主元分量,并按抽取的主元分量进行图像压缩.实验结果表明:该方法在保证较高压缩比的前提下,具有很好的峰值信噪比值与视觉效果;另外当取的较少图像主元分量压缩后,其重构图像仍能保持较好的清晰度,与基于四元数奇异分解的压缩方法相比较,克服了重构图像出现的纵横纹理性模糊;实验还表明该代数框架下的网络具有很好的泛化能力,其权值矩阵可直接应用于其他RGB图像的压缩,无需再次训练,具有压缩普适性.
This paper proposed a method for RGB color image compression based on Young-Helmholtz three-color theory after researching photosensitive mechanism proposed by Yong.Firstly,the GHA-NN was constructed under the framework of Young-Helmholtz algebra,the principle components(PCs) of the RGB image were extracted,and the image was compressed by the extracted CPs.The experimental results indicated that on the premise of guarantee high compression ratio,it had reasonable peak value of Signal-to-Noise Ratio(PSRN),and the visual effect was clear;in addition,when selected fewer PCs of the image for extracting based the proposed method,the results still could keep the reconstructed image clarity,and got over the textured blurred deficiency of reconstructed image which was compressed by Quaternion SVD.Moreover the network had strong generalization ability,the weight-matrix could be applied directly to compress other RGB image without training,so the weight-matrix had its some degree of universality for image compression.
出处
《小型微型计算机系统》
CSCD
北大核心
2013年第3期677-679,共3页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(60871093)资助
作者简介
丁立军。男，1979年生，博士研究生，研究方向为图像处理与模式识别；
冯浩，男，1956年生，博士，教授，博士生导师，研究方向为人工神经网络、模式识别；
华亮。男，1979年生，博士研究生，副教授，研究方向为机器视觉与模式识别．
