Abstract: In order to solve the problem of insufficient information provided by a single feature in complex scene recognition, a multi-scale census transform of difference of distant neighbors feature is proposed. Gabor filtering is used to obtain the multi-scale feature information of the image, and the improved census transform of difference of distant neighbors feature is extracted from the original image and the Gabor filtered image to generate a multi-scale descriptor. Finally, the multi-scale census transform features and bag-of-visual-word (BOVW) features are merged by kernel collaborative representation to classify scenes. This method considers the scale information and the long distance point information, and solves the problem of low discrimination of a single feature. The algorithm is verified on two standard datasets, and the experimental results show that the recognition rate of the proposed method is better.

Key words: scene classification, multi-scale, statistical features, collaborative representation

摘要： 针对复杂场景识别中单一特征提供信息不充分这一问题情况，提出一种多尺度远距离点差值统计变换特征。通过Gabor滤波获得图像的多尺度特征信息，在像素图和滤波图上分别提取改进的远距离点差值统计变换特征，从而生成多尺度描述子，最后将多尺度统计变换特征和视觉词袋模型特征通过核协同表示融合后进行场景分类。该方法充分考虑了尺度信息和远距离点信息，解决了单一特征区分度低的问题。算法在两个标准数据集上进行对比实验，结果表明所提算法取得了较好的识别效果。

关键词: 场景分类, 多尺度, 统计特征, 协同表示