摘要局部特征描述子在图像识别过程中起着举足轻重的作用。本文主要介绍了尺度不变特征变换算法(简称SIFT),方向梯度直方图算法(简称 HOG),方向纹理曲线算法(简称OTC)等局部特征描述子方法;同时也介绍了词包模型(BOW) ,空间金字塔模型(SPM)等特征编码方法以及支持向量机分类器。实现了图像识别的框架,并在此框架中完成了几种算法在不同图像数据库上的分类实验,比较了 SIFT,HOG,OTC 在不同数据库、不同实验参数、不同编码方式下的识别性能。本文用 Matlab 实现了OTC 局部特征描述子,针对 OTC 特征描述子的特征可视化部分为 H-bin 添加了一个权重,使之能适应不同的输入图像。为了提升OTC 特征的计算效率,我们利用 PCA对OTC每个局部区域的高文特征进行了数约减。 26041 毕业论文关键词 局部特征描述子 主成分分析 特征编码 支持向量机 图像识别
Title Local Descriptors and Applications in Image Recognition
Abstract
A fine local feature descriptor plays a significant role in image recognition.
This paper firstly introduces the Scale Invariant Feature Transform, Histogram
of Oriented Gradient, and Oriented Texture Curves. Then the feature coding models
including the Bag of Words and its extension to the Spatial Pyramid Matching are
described. Finally , SVM is introduced and employed for classification. We
implement the framework of image recognition and report the performances of SIFT,
HOG, OTC with several parameters and feature coding schemes in different image
databases. We re-implement OTC method in Matlab and add a weighted value for H-bin
in order to achieve a better visual performance of OTC features. PCA is utilized
to reduce the high feature dimension of each image patch.
Keywords local descriptor PCA feature coding SVM images recognition