基于小波矩的车辆特征提取算法研究

Research on Vehicle Feature Extraction Algorithm Based on Wavelet Moment. Computer Engineering and Applications

摘要:

车辆目标图像特征提取是智能交通系统中车辆识别与分类的关键问题。在车型提取算法中，矩特征是较为常用的车型特征描述子。针对Hu矩的七个特征分量在数量级上差别较大且受比例因子影响的问题，基于不变矩和小波能量的原理和特点的研究，重点提出了基于小波矩的特征提取算法，并应用于车辆的特征提取。最后的实验对实际车辆图像进行采集，对预处理图像进行小波分解得到三级子图像，对子图像求取修正Hu不变矩，将不变矩作为特征量，利用最小邻距离分类得出识别结果。最后的实验结果显示，通过这种方法提取的特征量具有平移、旋转、比例不变性，能反映目标图像的重要的、本原的属性，与传统Hu矩相比，识别率提高了13.5%，达到了预期的目标。

Abstract:

Vehicle image features extraction is a key problem for vehicle recognition and classification in the intelligent transportation system. The moment feature is often used as feature descriptors on vehicle types in the vehicle types extraction algorithm. Aiming at the great quantitative difference among the seven feature components of Hu moment and the disturbance from coefficient scale on them. Propose the feature extraction algorithm based on wavelet moment, lying on the principle of invariant moment and wavelet-energy. And the algorithm is applied on the vehicle image feature extraction. The real vehicle images are collected and pretreated in the experiment. After wavelet decomposition for the pretreatment images, the three-stage-subimages can be obtained, whose modified Hu moment also can be acquired through calculation. The modified moment is regarded as the feature. The result shows that the features extracted by the algorithm have the ability of keeping invariant after translation, rotation and scale transformation, which also could reflect the vital and essential attributes of target images and achieve the goals expected. Compared with the traditional Hu moments, recognition rate increased by 13.5%