一种结合显著性检测与词袋模型的目标识别方法  

An object recognition method combining saliency detection and bag of words model

作　　者：李伟生[1] 陈曦[1] LI Wei-sheng CHEN Xi(Chongqing Key Laboratory of Computational Intelligence,Chongqing University of Posts and Telecommunications,Chongqing 400065,Chin)

机构地区：[1]重庆邮电大学计算智能重庆市重点实验室,重庆400065

出　　处：《计算机工程与科学》2017年第9期1706-1713,共8页Computer Engineering & Science

基　　金：国家自然科学基金(61272195;61472055);重庆市基础与前沿研究项目(cstc2014jcyjjq40001)

摘　　要：针对词袋模型易受到无关的背景视觉噪音干扰的问题,提出了一种结合显著性检测与词袋模型的目标识别方法。首先,联合基于图论的视觉显著性算法与一种全分辨率视觉显著性算法,自适应地从原始图像中获取感兴趣区域。两种视觉显著性算法的联合可以提高获取的前景目标的完整性。然后,使用尺度不变特征变换描述子从感兴趣区域中提取特征向量,并通过密度峰值聚类算法对特征向量进行聚类,生成视觉字典直方图。最后,利用支持向量机对目标进行识别。在PASCAL VOC 2007和MSRC-21数据库上的实验结果表明,该方法相比同类方法可以有效地提高目标识别性能。Given that the bag of words model is quite sensitive to background noise and that visual words in the background are not relevant to objects, we propose an object recognition method which combines saliency detection with the bag of words model. Firstly, the region of interest from the original image is adaptively gained by using the graph-based visual saliency （GBVS） algorithm and the AC algorithm. The combination of the two detection methods can avoid incomplete region of interest. Secondly, we extract local features from the region of interest by using the scale invariant feature transform （SIFT） descriptor. Then, we use the peak density clustering algorithm to classify the features and generate a visual dictionary histogram by clustering local features. Finally, we employ the support vector machine （SVM） classifier to classify and recognize objects. Experiments on PASCAL 2007 and MSRC-21 databases verify the effectiveness of this method. Experimental results show that the proposed method can effectively improve the performance of object recognition.

关 键 词：词袋模型 显著性检测 密度峰值聚类 目标识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]