分层纹理特征和梯度特征融合的图像分类

Image Classification Based on Fusion of Layered Texture and Gradients Feature

为了提高分类检索和识别的准确率,提出了分层纹理特征和梯度特征融合的方法,即分层中心对称局部二值模式(CS-LBP)和梯度方向直方图(HOG)的特征融合方法。首先,对原始图像进行多次CS-LBP特征的提取,得到3层不同的特征图像;然后对特征图像进行大小相等、不重叠分块,分别提取每块CS-LBP特征和HOG特征,形成每一层的特征;再将特征图像的特征进行融合。分别在标准图像库和人脸库上进行仿真,研究结果表明:提出的分层融合方法的分类查准率和识别率比传统方法分别提高了15%和10%。

In order to improve the accuracy of image classification retrieval and recognition,a new method was proposed,which is the fusion of layered texture feature and gradients feature,that is the layered center symmetric local binary pattern( CS-LBP) and histogram of oriented gradients( HOG). The CS-LBP was utilized to extract the three level features from the original image and get different feature images. Then every layer feature images were divided into blocks with the same size. The two different histogram features which are the CS-LBP and HOG of every layer feature image were formed. The multilevel features were obtained and different level features were fused to represent original image. Experimental results based on the standard image database and face database demonstrate that the proposed approach increases the retrieval and recognition rates by 15% and 10% respectively than traditional methods.