Image Classification via fusing the latent deep CNN feature

Abstract

In recent years, the convolutional neural network (CNN) has made great achievements in image classification. It can extract features of image and classify them from a large number of image data automatically. Compared with these traditional feature extraction techniques (e.g., SIFT, HOG, GIST), the convolutional neural network can make better performance and does not need hand designed image features. However, how to further enhance the algorithm performance is still a hot spot in academic research. Therefore, in this paper, we propose a method to fuse the latent features extracted from the middle layers in a CNN to train a more robust classifier. First, we utilize the pretrained CNN models by caffe to extract visual features of 3-layer.Then, we use the SVM classifier for 3-layer features respectively, and we get their trained classifier. Finally, we combine three pretrained classifier into a classifier which is compared with 3-layer SVM classifier. The experiment is performed on Caltech-256 datasets. The experimental result shows that the combined classifier obtains good performance compared with the conventional CNN.