Thai Text Detection and Classification Using Convolutional Neural Network

Many foreign people don't know Thai language and most of the time Thai sign images, posters or text images do not have subtitles in English so, it is very necessary to have a system that can translate Thai text to English. In this paper, MSER and convolutional neural network (CNN) have used to understand Thai text in English. Firstly, region of interest has localized from natural image which is some particular Thai text. Then text has extracted and fed to CNN. We used a 7-layer self-designed CNN model that provides the output with an accuracy of 98%. The proposed system takes natural scene image as input and uses MSER, geometrical properties as well as bounding box algorithm to localize the text area then selected localized areas have fed to CNN and provide an output that has the English meaning for the Thai text image. This paper introduces a new approach of text translation by using image classification method. The proposed system can work on particular inputs which are indoor sign Thai text images.