一种改进的小波特征提取算法及其应用

Improved Algorithm of Wavelet Feature Extraction and its Application

特征提取是模式识别中的一个关键问题.为解决现有的基于灰度空间和梯度方向的小波特征用于目标物分类检测时对光照及背景噪声敏感的问题,提出一种改进的小波特征提取算法,即对感兴趣区域(Region of Interest,ROI)基于HSV颜色模型的V通道分量进行小波金字塔式分解,然后取塔式分解得到的小波系数幅值,对其进行归一化处理,最后进行阈值化处理.将改进的算法应用于基于单目视觉的静态图像后方车辆检测系统中,实验结果表明其能显著提高车辆识别效果,增强系统的鲁棒型.

Feature extraction is a key point in pattern recognition field. Currently, the wavelet features based on gray space and gradient orientation are sensitive to the illumination changes and background noise contained to the vehicle region. In order to deal with this problem, an improved algorithm of wavelet feature extraction is proposed. In particular, wavelet pyramid decomposition is performed, which is based on the V channel of the HSV color model of the ROI （Region of Interest）, after that the coefficient magnitudes are obtained and then they are scaled, finally the threshold process is performed on the sealed data. With the application in a rear-vehicle detection system for static image based on monocular vision, the experimental results show the significant improvements both in vehicle detection and robustness.