Image Classification Based on Image Hash Convolution Neural Network

Abstract

In image classification tasks, in order to improve the classification accuracy, we need to extract the multidimensional sample characteristics. However, for massive amounts of data, calculation and storage is a big bottleneck. In this paper, a method named image hash was proposed to solve this problem. Image hash can code high-dimensional image feature for simple binary code. Feature extraction is the most important step in image hash. In the current works about image hash, feature extraction needs artificial experience to design feature extractor, which is complicated and not intelligent. Convolution neural network can take original image as input to obtain from the bottom level to the top level of characteristics, which is robust for translation zooming and rotation etc. Therefore, this paper proposes the image hash combined with convolution neural network for image classification, and the experiment proves that it has good classification effect.