词袋模型中梯度方向离散精度阈值经验分析  

Empirical analysis of threshold for gradient direction discrete precision in bag-of-words model

作　　者：生海迪 段会川[1,2] 孔超[1,2] 

机构地区：[1]山东师范大学信息科学与工程学院,山东济南250014 [2]山东师范大学山东省分布式计算机软件新技术重点实验室,山东济南250014

出　　处：《计算机工程与设计》2014年第9期3270-3273,共4页Computer Engineering and Design

基　　金：国家自然科学基金项目(61272094;61373149)

摘　　要：为避免常规构建词袋模型在选取梯度方向离散精度时存在的盲目性,以及解决不恰当离散精度构造出的词袋模型分类率不高的问题,研究在词袋模型中提取图像特征时梯度方向离散精度是否存在明显的阈值,使选取该阈值时能构造出最佳的词袋模型。基于面向稠密特征提取可快速计算的局部图像特征描述,选择不同的梯度方向离散精度分别进行大量实验,实验结果表明,存在一个明显、统一的阈值24,选择该阈值构造的词袋模型能够得到最高正确分类率。To avoid the blindness of selecting the gradient direction discrete precision in routinely building bag-of-words model, as well as low classification precision results gotten from using poor bag-of-words model, whether there was an obvious threshold for gradient direction discrete precision during the image feature extracting process in the bag-of-words model was studied to obtain a threshold which led to the best bag-of-words model. Based on fast local descriptor oriented dense feature extraction, a large number of experiments were done with different gradient direction discrete precisions respectively. Experimental results show that 24 is the clear and unified threshold, the bag-of-words model which chooses the threshold can get the highest eorrect classification results.

关 键 词：图像分类 词袋模型 DAISY描述子 梯度方向 梯度方向离散精度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]