Preprocessing Method for Training Dataset in Character Recognition using Convolutional Neural Network
Taku Akase, Hitoshi Nakao, Lifeng Zhang · 2019
Nowadays, image processing technology is used in the logistic industry widely to eliminate a labor shortage and prevent human errors. However, the existing image processing technology cannot be applied on some work processes such as to identify the product name written in characters that printed on an unstable package surface, because the color and font aren’t uniform and the shapes are deformed always. Therefore, we propose a system previously which enabled this task using character recognition technology by Convolutional Neural Network (CNN) with a certain accuracy. In this research, we focus on how to make an optimal training set which can obtains a high accuracy. As a result, the accuracy is improved to 99.8% by adding the diversity of the image's background for learning process.