基于改进BOF算法的图像识别和分类

Image recognition and classification based on improved BOF algorithm

对bag of features(BOF)算法进行研究与改进,并将其应用到图像识别和分类中。针对传统BOF算法执行效率低以及分类精度不够高等缺陷,提出一种结合SURF(speeded up robust feature)与空间金字塔匹配原理的优化方法相结合的图像识别与分类算法。SURF算法可提高执行效率,而空间金字塔匹配原理的优化方法可提高分类精度。首先对分类图像应用SURF算法提取特征描述符并生成视觉词典,该算法提取的视觉词典能更有效地表示图像特征,且能应对多变的尺度;然后应用空间金字塔匹配原理对图像采用视觉词典的直方图表示,进一步提高分类的准确度;最后利用LIBSVM分类器进行分类。在Graz,Caltech-256和Pascal VOC 2012这3个数据集中进行实验测试。研究结果表明:该方法与传统的BOF算法相比提高了执行效率和分类精度。在数据实验中通过与近几年一些相关研究工作在分类准确率方面进行对比,该方法具有很大的优越性。

Improved bag of features（BOF） algorithm was applied to image recognition and classification. In view of the low efficiency and low classification accuracy of the traditional BOF algorithm, a new recognition and classification algorithm combined SURF（speeded up robust feature） with spatial pyramid matching principle was proposed. SURF algorithm can improve the efficiency, and spatial pyramid matching principle can improve the classification accuracy. Firstly, the image feature was extracted by SURF algorithm and the codebook was generated using the features which were able to respond the changing scales. Secondly, the spatial pyramid matching principle was applied to the image histogram＇s codebook which can improve the accuracy of the classification. Finally, the image histogram＇s codebook was used to be the input of LIBSVM classifier. The experiments were carried out based on Graz, Caltech-256 and Pascal VOC 2012. The results show that the proposed method is better than the traditional method in the efficiency and classification accuracy. In addition, the proposed method is compared with some related research work in classification accuracy, and the proposed method has obvious advantages.