Gabor Feature Based Convolutional Neural Network for Object Recognition in Natural Scene

Feature extraction and classification are two important components in object recognition. While the traditional methods design these components individually, the deep neural networks jointly learn these two parts. In this paper, we propose a method of the convolutional neural network combined with Gabor filters for strengthening the learning of texture information. We called this model as Gabor-CNN below. Through experiments, the approach achieves the recognition rate of 81.53%, yielding a 1.26% promotion in the average accuracy rate compared with the results obtained using the convolutional neural network model alone on the ImageNet10 dataset, as well as significantly outperforming the traditional method based on Bag-of-Words model with SIFT.