LAUSR

A significant issue in the domain of optical character recognition is handwritten text recognition. Here, two novel feature extraction techniques are proposed using a fixed-size sliding window, and also an edit distance-based architecture is suggested to recognize the off-line characters. These feature extraction techniques are designed for text recognition from the text images. It’s an off-line approach, that is why data from scanned documents or natural scenes are taken as input. In this paper, the freely available datasets, known as Chars74k and MNIST for English alphabets and digits are used. The proposed feature extraction technique for the off-line text images of characters as well as numbers generates the features successfully. The impact of the proposed method on text recognition accuracy is computed using several state-of-the-art machine learning algorithms. After that, these are again compared with the proposed Edit distance based text recognition system with the help of different conducted experiments. The proposed model has reached an accuracy of more than 96% for the MNIST dataset.