Recognition of Various Handwritten Indian Numerals Using Artificial Neural Network

Abstract

In current years, extracting documents written by hand is extensively studied topic in image analysis and optical character recognition. These extractions of document images find their applications in document analysis, content analysis, document retrieval, and much more. Many complex text extracting processes such as maximization likelihood ratio (MLR), neural networks, edge point detection technique, corner point edge detection are generally employed for extraction of text documents from images. This article uses feed-forward propagation model of neural network for recognition of various Indian handwritten numerals like Punjabi, Hindi, Bengali, Telugu, and Marathi. Recognition is achieved by initially acquiring the image, then preprocessing it and then feature extraction. Preprocessing is performed by binarizing the image and segmenting the preprocessed image by cropping it to its edges. Feature extraction involves the normalizing the numeral matrix into 12 × 10 matrixes. Feature recognition applies artificial neural network for detection of numerals. The network is constructed with 120 input nodes, 10 hidden layer nodes, and 10 output nodes. The network has one input, single output, and a hidden layer. The numbers used for training are divided using a morphological method, and the network is trained for various Indian numerals. The proposed system has 98% recognition accuracy with respect to training data.