基于纹理特征与BP神经网络的运动车辆识别

在Gabor小波滤波器组与图像卷积值作为特征向量达到很高识别率的基础上,提出了一种特征值加权的Gabor小波纹理特征的提取方法.首先Gabor小波函数与纹理图像做卷积,然后加权处理尺度各不相同和方向各不相同的的卷积值,最后将均值和方差看作它们的特征向量,该方法使特征维数有所降低,并利用BP神经网络进行训练和仿真,实现运动车辆纹理图像的自动分类,达到运动图像的识别.实验结果表明此算法有效降低了图像的识别错误,增强了稳健性,对质量差的图像能够有效识别.

The motor vehicle identification based on Texture feature and BP neural network

On the basis of the Gabor wavelet filter group and the image convolution values as the feature vector can achieve a high recognition rate, a feature-weighted method of extracting texture has been proposed. First, Gabor wavelet function and texture image deconvolution, then we will extract the convolution values in different scales and different directions. After making the weighting process, taking its mean and variance as the characteristic vector, which greatly reduces the feature dimension.Finally, BP neural network is used to making training and simulation, in order to achieving the automatic classification of texture images of moving vehicles and the identification of moving images. The experimental results show that this algorithm can effectively reduce the recognition error of the image and enhance the robustness. To the poor quality images, It can make the effective recognition.