一种多尺度时频纹理特征融合的场景分类算法

Multi-scale time-frequency texture feature fusion algorithm for scene classification

场景分类目前是机器视觉领域的一个研究热点,为了解决该研究领域中分类特征的提取问题,提出了一种多尺度纹理描述子(MSTD)特征。首先,采用小波变换,获得图像在时频域上的多尺度纹理视觉全局特征信息,之后提取反映局部细节的局部二值模式(LBP)特征,在时频域上进行融合,生成多尺度纹理描述子特征,以此作为图像分类的依据,最后采用支持向量机(SVM)作为分类器进行场景分类。在4个标准数据集上进行测试,实验结果表明,该方法具有较高的分类正确率,对室外场景的分类正确率都在84%以上。所提出的分类方法充分考虑了全局特征和尺度信息,增强了单层特征的区分度,有效地改善了分类的精度。

Scene classification is currently a hot topic in the field of machine vision.In order to solve the problem of classification feature extraction,a multi-scale texture descriptor （Multi-scale Texture Descriptor,MSTD）characteristic is presented.Firstly,the wavelet transform is used to obtain a multi-scale global visual texture feature information in the time-frequency domain.The LBP feature reflecting local details and fusing in the time-frequency domain is extracted to generate multi-scale texture descriptors features.The final SVMis used as a classifier to classify the scene.Based on four standard data sets,experimental results show that the presented method has higher classification accuracy,and the outdoor scene classification accuracy rate is above 84%.The proposed classification method takes into account the global features and scale information to enhance the discrimination of single characteristics,which effectively improves the accuracy of the classification.