一种基于二进制分辨矩阵的视觉单词约简方法  

A Visual Word Reduction Method Based on Binary Discernibility Matrix

作　　者：王俊伟 张素兰[1] 胡立华[1] 

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《太原科技大学学报》2017年第5期359-364,共6页Journal of Taiyuan University of Science and Technology

基　　金：校博士启动基金项目(20132005;20162009)

摘　　要：针对图像场景生成视觉词典过程中产生冗余视觉单词而导致分类误差的问题,提出了一种基于二进制分辨矩阵的视觉单词约简方法。该方法首先通过调整归一化阈值α的取值,生成关于训练图像初始视觉词典不同的0-1信息决策表和二进制分辨矩阵,然后以二进制分辨矩阵行列方向1的个数作为启发信息,将二进制分辨矩阵行方向上只有一个1的视觉单词作为核视觉单词,列方向上1出现总数最大的视觉单词作为重要视觉单词,并将这些视觉单词构成的集合作为描述图像分类的决策规则。最后采用OT8作为数据集,实验验证了该方法能够有效减少冗余视觉单词对图像场景分类的影响,提高图像场景分类的精度。Aiming at the problem of classification error from redundant visual words,which is generated during the process of visual dictionary,a reduction method of visual words is presented based on binary discernibility matrix.First,different 0-1 information decision tables and binary discernibility matrix from initial visual dictionary are adjusted through adjusting the normalized threshold value of α. Then the number of 1＇s in binary discernibility matrix is regarded as the heuristic information. The visual words with only one number of 1 in the binary discernibility matrix＇s line direction are regarded as nuclear Visual words,the ones in the column direction as important visual words,and these words are combined to construct the classification decision rules for the description of image. In the end,experiments show that the method can effectively reduce the influence of redundant visual effects of words on scene classification and improve the accuracy of scene classification by taking OT8 as data sets.

关 键 词：视觉词典 二进制分辨矩阵 归一化阈值 冗余视觉单词 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]