Convolutional Neural Networks for Scene Image Recognition

Yulian Li, Chao Luo, Hao Yang, Tao Wu

Abstract: Words are the most indispensable information in human life. It is important to analyze and understand the meaning of words. Compared with the general visual elements, the text conveys rich and high-level meaning information, which enables the computer to better understand the semantic content of the text. With the rapid development of computer technology, the research on text information recognition has made great achievements. However, in the face of dealing with text characters in natural scenes, there are certain limitations in the recognition of natural scene images. Because scene images have more interference and complexity than text, these factors make the identification of natural scene image texts facing many challenges. This paper focused on the recognition of natural scene image texts, and mainly studied a text recognition method based on deep learning in natural scene images. Firstly, text recognition is based on Kares using the Dense Convolutional Network (DenseNet) network model by using the existing standard test data set. Secondly, each character is classified using Softmax outputs to achieve the use of automatic learning. The characteristics of the context replace the manually defined features, which improve the recognition efficiency and accuracy. Lastly, the text recognition of the natural scene image is realized. And the method is suitable for problems encountered in medical images.