基于Bag of Features算法的车辆图像识别研究  

Vehicle Image Recognition Study Based on Bag of Features

作　　者：何友松[1] 吴炜[1] 陈默[1] 杨晓敏[1] 罗代升[1] 

机构地区：[1]四川大学电子信息学院图像信息研究所,四川成都610065

出　　处：《电视技术》2009年第12期104-107,共4页Video Engineering

摘　　要：将Bag of Features算法引入汽车图像识别领域中,并提出了将DoG(Difference of Gaussian)特征提取算法和PLSA分类算法结合在一起实现车辆和背景图像分类。首先用DoG特征提取算法提取图像特征,用这些特征聚类产生码书并对图像进行柱状图描述,最后设计PLSA分类器对车辆图像和背景图像进行分类。实验对比了该算法与Tamura纹理特征算法和Gabor纹理特征算法在车辆图像识别中的效果。结果表明本文算法分类正确率优于另外两种方法。Bag of Features is introduced to the field of vehicle image recognition, and a way is proposed to combine DoG features extraction algorithm with PLSA（Probabilistic Latent Semantic Analysis） to classify vehicle images and background images. Image features with DoG is extracted firstly, and then codebook via clustering is generated, at last images with PLSA is classified. In the experiment, the results of vehicle images recognition by the proposed method are compared with the results of Tamura texture and Gabor texture respectively. Experimental results show that this method has better performance than the other two algorithms.

关 键 词：BAG of Features算法 码书 SIFT K-MEANS 概率潜在语义分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]