基于SVM和VAR/LBP的车脸识别

Vehicle Make and Model Recognition Based on SVM and VAR/LBP

局部二值模式(LBP)作为一种纹理特征在图像识别方面广泛应用并取得了良好的效果,针对原始的LBP特征统计直方图特征维数相对较高并且不具有旋转不变性的问题,采用对图像的旋转与光照具有较好鲁棒性的VAR/LBP特征,并将该方法应用到车脸图像的识别;对车脸图像提取的VAR特征图像进行分块,并分别统计子块图像的LBP直方图特征,并结合SVM模型训练出了分类效果较好的分类器。通过实验证明,VAR/LBP与SVM相结合的算法对车脸图像识别可达到98.6%的正确率,具有良好的鲁棒性。

Local Binary Pattern（ LBP） has widely used in image recognition with extracting texture feature,for the high statistical histogram dimensions of LBP and unable to have rotation invariance,so the VAR/LBP feature which is robust to the light and rotation is applied to vehicle make and model recognition. The vehicle image is extracted texture feature with blocking statistical histogram,a better classifier is trained based on the statistical histogram feature and SVM. Experiments show that the combination of SVM and rotation and uniform invariant LBP can achieve 98. 6% correct rate,have a good robustness.