|
|
|
| A dimension reduction method of the histogram of oriented gradients |
| FU Hongpu1,2, ZOU Beiji1 |
1. Ministry of Education-China Mobile Joint Laboratory for Mobile Health, School of Information Science and Engineering, Central South University, Changsha 410083, China;
2. Department of Information Science and Engineering, Hunan First Normal University, Changsha 410205, China |
|
|
|
Abstract To characterize the local object appearance and shape, histograms of oriented gradients (HOG) divide an image window into small spatial regions (cells), and accumulate a local 1-D histogram of gradient directions over the pixels of the cell. The normalized combined histogram entering of a larger spatial region (blocks, are consisted of several cells) forms the representation. In order to weaken regional quantization aliasing, blocks are partly overlapped when HOG is computed in detection windows. Yet, it will increase the dimension vastly. So, it will bring extra computation for object detection application. By expanding the area (spatial region) where pixel gradients are interpolated between neighbor cells' centers, and setting the scale of block Gaussian weights properly, the overlapped area between blocks is cancelled. Then, the dimension of HOG feature in a 64×128 detection window reduces from 3 780 to 1 152, and region quantized errors are decreased. Experiment results on INRIA pedestrian dataset show that the performance of the 1 152-dimensional HOG and that of the original HOG are almost the same, however, its detecting speed is significantly improved.
|
|
Received: 25 July 2016
Published: 08 July 2017
|
|
|
一种方向梯度直方图的降维方法
为描述对象的局部外观和形状，方向梯度直方图首先将图像划分成小区域（被称为cell），然后在其上累加像素梯度方向的一维直方图.在被称为block的较大区域（由数个相邻的cell组成）上连接cell的直方图，经归一化处理形成特征向量.为减弱由block引起的区域量化走样，在计算检测窗口的特征时，采取部分重叠block的措施，从而大大增加了特征维度以及目标检测时的计算量.通过扩大参与相邻cell之间像素梯度插值的面积，并设置适当的高斯平滑核尺度，可消除block重叠，从而将64×128尺寸的窗口的方向梯度直方图特征维度由3 780降低为1 152.INRIA的行人数据集实验表明，该方法也可减弱区域量化走样，且其性能与原方向梯度直方图几乎相当，而检测速度却显著提高.
关键词：
方向梯度直方图,
降维,
线性插值,
重叠采样,
行人检测
|
|
[1] DALAL N, TRIGGS B. Histograms of oriented gradients for Human detection[C]//The IEEE Conference on Computer and Pattern Recognition.Washington D C:IEEE Computer Society,2005:886-893.
[2] DOERSCH C, EFROS A. Improving the HOG descriptor[J/OL] [2016-04-09]. http://www.cs.cmu.edu/~cdoersch/projects/hogimprove/hogimprove.pdf.
[3] QUATTONI A, CARRERAS X,TORRALBA A. A latent variable ranking model for content-based retrieval[C]//34th European Conference on Information Retrieval.Berlin:Springer,2012.
[4] AVIDAN S, SHAMIR A. Seam carving for content-aware image resizing[C]//Proceeding of ACM SIGGRAPH. New York:ACM Transactions on Graphics,2007.
[5] BOSH A, ZISSERMAN A, MUNOZ X. Image classification using random forests and ferns[C]//IEEE International Conference on Computer Vision. Rio de Janeiro:IEEE Computer Society,2007:1-8.
[6] NAM W, DOLL'AR P, HAN J H. Local decorrelation for improved pedestrian detection[C]//NIPS2014 Montreal:Advances in Neural Information Processing Systems,2014:424-432.
[7] YANG B, YAN J, LEI Z, et al. Convolutional channel features[C]//ICCV2015. Santiago:Computer Science,2015:82-90.
[8] CAI Z, SABERIAN M, VASCONCELOS N. Learning complexity-Aware cascades for deep pedestrian detection[C]//ICCV2015. Santiago:Computer Science,2015:3361-3369.
[9] DANG L, BUI B, VO P D, et al. Improved HOG Descriptors[C]//In the Third International Conference on Knowledge and Systems Engineering. Washington DC:IEEE Computer Society,2011:186-189.
[10] DOLLAR P, TU Z, PERONA P, et al. Integral channel features[C]//BMVC. Belongie:BMVC 2009 London England,2009:1-11.
[11] FELZENSZWALB P F, GIRSHICK R B, MCALLESTER D, et al. Object detection with discriminatively trained part-based models[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,2010,32(9):1627-1645.
[12] FRYER J, MCINTOSH K. Enhancement of image resolution in digital photogrammetry[J].Photogrammetric Engineering & Remote Sensing, 2001, 67(6):741-749.
[13] VIOLA P, JONES M J. Robust real-time face detection[J]. International Journal of Computer Vision, 2004, 57(20):137-154.
[14] CHANG C, LIN C. LIBSVM:A library for support vector machines[J]. ACM Transactions on Intelligent Systems and Technology, 2011, 2(3):1-27. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm |
|
Viewed |
|
|
|
Full text
|
|
|
|
|
Abstract
|
|
|
|
|
Cited |
|
|
|
|
| |
Shared |
|
|
|
|
| |
Discussed |
|
|
|
|
