融合多尺度码本的全局编码图像分类  

Image classification based on global coding combined with multi-scale codebook

作　　者：董振宇[1] 赵杰煜[1] 祝军[1] 

机构地区：[1]宁波大学信息科学与工程学院,宁波315211

出　　处：《中国图象图形学报》2015年第2期183-192,共10页Journal of Image and Graphics

基　　金：科技部国际科技合作专项(2013DFG12810);国家"十二五"科技支撑计划项目(2012BAF12B11);浙江省国际科技合作专项(2013C24027)

摘　　要：目的词袋模型在图像分类领域中的分类效果主要受限于局部特征的量化误差。针对这一点,提出一种融合多尺度码本的全局编码图像分类方法,有效减少特征量化误差。方法通过使用多尺度特征密集采样,构建多尺度码本,使码本具备一种层次结构,通过充分利用图像特征的流形结构,计算码本全局信息,实现全局编码。通过本文方法得到的编码系数比较平滑和准确。最后使用多路径方法,分别将不同尺度的特征表示进行级联,得到最终的图像特征表示。这种特征表示具备了一定程度上的尺度不变性。结果在UIUC-8和Caltech-101两个常用的标准图像数据集上进行测试,分类准确率分别达到88.0%和83.2%。结论实验结果表明,相比于基于固定尺度码本的局部编码方法,本文方法在分类识别率方面有了显著提升。Objective The performance of the Bag-of-Words model in the field of image classification is limited mainly by the quantization error of the local feature. To reduce the quantization error of the local feature effectively, an image classifi- cation method based on global coding combined with multi-scale eodebook is proposed. Method A global coding is imple- mented by utilizing fully the manifold structure of the image features and by computing the global information of the code- book. The coding coefficients obtained by the method are relatively smooth and accurate. Furthermore, a multi-path method is designed to integrate all feature representations to describe the image. To a certain extent, this method can achieve the scale invariance of feature representations. Conclusion Several experiments are conducted on two commonly used bench- mark data sets, namely, UIUC-8 and Catltech-101, and the average classification accuracy rates reach up to 88.0% and 83.2% , respectively. Result Experimental results show that the proposed method improves the performance significantly compared with the fixed-scale locality-constrained coding methods.

关 键 词：图像分类 特征编码 多尺度码本 全局编码 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]