A Deep Convolutional Neural Network based Chinese Menu Recognition App

This paper presents a Deep Convolutional Neural Network (DCNN)-based Chinese Menu Recognition App. A DCNN is a multi-layer feed-forward neural network and is an alternative type of Deep Learning. In the proposed application, the network was constructed by: (1) Three convolutional layers, with 20, 50 and 5 convolutional kernels, separately, and (2) two max-pooling layers, and (3) a full-connected layer with 500 neurons. Users use camera to capture the dishes name from the Chinese menu and send to the server. The server side was developed by Python Flask framework and returns the translated English food name as well as the corresponding image. Compared to the Google Translate app, our app presents a better performance in dealing with high noise, low resolution, distorted and mixed topic images.