基于BoF和迹变换多特征融合的图像纹理分类研究

Research on image texture classification based on multi⁃feature fusion of BoF and trace transform

针对BoF模型缺少几何特征、结构特征的表达,对纹理图像特征描述不充分等问题,提出一种基于BoF和迹变换多特征融合的图像纹理分类方法。首先通过关键点检测的方法获取纹理图像的碎片化图像,然后提取碎片化图像的迹变换特征和SIFT特征,通过特征交叉编码的方式和动态鉴别能量的方法,获取迹变换特征和SIFT特征的融合特征并进行特征单词优选,再以BoF模型进行特征编码,最后输入到支持向量机(SVM)中进行训练、预测和分类。实验在OutexTC10/TC12000和KTHTIPS纹理数据集上分别取得了100%、99.87%和97.6%的识别精度,结果表明该设计方法对具有几何特征、结构特征的纹理图像可以获得较好的分类效果,有效地提高了纹理分类的识别性能。

In view of the lack of expression of geometric and structural features of BoF model,and insufficient description of image texture features,an image texture classification method based on multi⁃feature fusion of BoF and trace transform is proposed.The method of key point detection is used to obtain the fragmented image of the texture image and then extract the trace transformation feature and SIFT feature of the fragmented image.The methods of feature cross⁃coding and the dynamically identifying energy are adopted to obtain the fusion feature of the trace transform feature and the SIFT feature,and perform the feature word optimization.BoF model is taken to encode the fusion features,and finally input them into the support vector machine(SVM)for training,prediction and classification.The recognition accuracy of 100%,99.87%and 97.6%was achieved in an experiment on the OutexTC10/TC12000 and KTHTIPS texture datasets.The results show that the designed method can obtain a better classification effect of texture images with geometric and structural features,and effectively improve the recognition performance of texture classification.