摘要
为了更有效地提高图像分类性能和准确率,提出一种基于HOG-PCA的高效图像分类方法。首先通过提取方向梯度直方图(HOG)特征并作特征白化,再随机下采样进行尺度统一,随后采用主成分分析(PCA)进行特征映射,最后用最小二阶范数判定进行最近邻分类。实验中,采用C++,基于OpenCV和Darwin实现了提出的方法,并在Pascal 2012数据集上进行测试,比较了该方法和BOW-SVM方法的准确率和运行性能。实验证明,提出的方法具有更高的准确率和更好的运行性能。
This paper presented a novelty classification method based on histogram of gradient (HOG) features and principle component analysis (PCA). Firstly, it extracted HOG features of image and whitened the features. At the second stage, it subsampled the features to an uniform scale, then extracted the image HOG features and projected these features into PCA space. At the last stage, it compared each test image with the PCA projections of training images, searched the nearest image comparing to the test image, finally it decided which class the test image belonged to and its max confidence scores. It implemented this method using C++, and based on OpenCV 2. 4. 3 and Darwin 1. 3. 2 platform, tested this method on Pascal 2012 dataset with comparing to BOW-SVM method. The experiment shows this method has advantages both on running time and classification precision.
出处
《计算机应用研究》
CSCD
北大核心
2013年第11期3476-3479,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(60702071)
国家"973"计划基础研究项目(2010CB732501)
四川杰出青年基金资助项目(09ZQ026-035)
关键词
方向梯度直方图
主成分分析
最小二阶范数
图像分类
图像特征
HOG(histogram of oriented gradients) PCA(principal component analysis) least squares norm image classification image feature