An Analysis of Convolutional Neural Networks for Image Recognition

Abstract

At present, Convolution Neural Network (CNN) has been widely applied to image recognition. Most of the researches focus on the CNN algorithm research, but there are few studies on the impact of the CNN’s parameters. In view of this situation, this paper focuses on the influence of each parameter on the experimental results. Through a series of Handwritten-Font-Recognition experiments using Convolution Neural Network, the relations between these parameters (e.g. the depths of CNN, the step sizes) in these modes and the influence of different parameter setting are preliminary grasped, which provides a reference for the study of CNN. We aim to help readers understand the relevant work methods and ideas by these experiments.