基于自组织增量神经网络码书产生方法在图像分类中应用.docVIP

基于自组织增量神经网络码书产生方法在图像分类中应用 　　文章编号2013 doi：10.11772/j.issn2013.07.1976 摘 要： 针对基于码书模型的图像分类方法忽略图像的拓扑信息及增量学习导致分类精度有限的问题，提出了基于自组织增量神经网络（SOINN）的码书产生方法。首先回顾了常见的码书编码方式；其次改进了基本的码书模型，利用SOINN自动产生聚类数目和保留数据拓扑结构的两项能力，寻找更有效的单词和设计更有效的编码方式，产生更合适的码书。实验结果显示在不同样本数和不同规模码书下分类精确度相对同类算法有最高将近1%的提升。该结果表明基于SOINN的码书产生方法显著提高了图像分类算法的精度，该方法还可以更高效、更准确地运用于各种图像分类任务。 关键词： 码书；图像分类；空间金字塔；SOINN 中图分类号： TP311 文献标志码：A 英文标题 Codebook generation based on selforganizing incremental neural network for image classification Combination of SOINN and codebook technology for image classification 英文作者名 YUAN Feiyun* 英文地址（ College of Information Engineering，Yulin University，Yulin Shaanxi 719000，China 英文摘要） Abstract： To solve the problem of ignoring topological information in incremental learning in traditional image classification， a new codebook generation method was proposed to improve the accuracy of image classification. After reviewing several codebook methods， the detailed method was discussed. Based on the SelfOrganizing Incremental Neural Network （SOINN） which can automatically generate clusters while conserving topological structures， the method produced a more effective way for representing words and coding. The experimental results show that the new method has at most nearly 1% precision increase over other similar algorithms in different scale of samples as well as different kind of codebook models. The results reveal that the new method has more appropriate and more accurate classifications for images. Also， it can be widely used in all kinds of image classification tasks with higher precision and efficiency. To solve the problem of ignoring topological information in incremental learning in traditional image classification，A new codebook generation method was presented to improve the accuracy of image classification. After reviewing several coodbook methods， the detailed method was discussed. Based on the SOINN which can