Published Paper Details:

With the rapid growth of document storage in modern society, handwritten character recognition has become an important research branch in the field of machine learning and pattern recognition. Meanwhile, how to effectively extract and store the information from documents plays a critical role in the document digitization system, especially in terms of ancient literature storage. However, handwritten characters vary with the writers and writing materials, even handwritten texts with the same content may have many differences, which brings challenges to the establishment of a high reliability recognition system. This project seeks to classify an individual handwritten word so that handwritten text can be translated to a digital form. We used two main approaches to accomplish this task: classifying words directly and character segmentation. For the former, we use Convolution Neural Network (CNN) with various architectures to train a model that can accurately classify words. For the latter, we use Long Short Term Memory networks (LSTM) with convolution to construct bounding boxes for each character. We then pass the segmented characters to a CNN for classification, and then reconstruct each word according to the results of classification and segmentation. This work aims to implement a segmentation strategy in the context of handwritten multi-digit string recognition. The LSTM model is built to be evaluated and benchmarked, also research about the effect of the different training sets on model performance is carried out. Finally, the text can be converted to digits, alphabet, and Tamil characters.