Online Handwritten Gujarati Word Recognition

Background

A lot of researchers have started working on offline handwritten character recognition after OCR. In offline handwritten character recognition system, handwritten documents are scanned and further processed to recognize written characters. For offline handwritten character recognition, remarkable work has been done for many languages.

Patel and Desai (2010) have proposed word segmentation of text lines into words for handwritten Gujarati text. The authors have used a projection profile with morphological operations for segmentation of lines into words. Desai (2010a) has proposed a Handwritten Gujarati numeral recognition system. He has used four profile vector-based features. He has used a feed-forward neural network for classification. He has achieved an accuracy of 82%. Desai (2010b) has proposed a Handwritten Gujarati numeral recognition system. He has used k-NN classifier with Euclidean distance method for classification. He has used a hybrid feature set which includes a subdivision of skeletonized image and aspect ratio as a statistical feature. He has achieved an accuracy of 96.99%. Patel and Desai (2011) have proposed zone identification methods for Gujarati handwritten word. They have used distance transform for identifying the upper, middle, and lower zone of a word. Desai (2012) has proposed character segmentation methods for old typewritten Gujarati documents. He has used radon transform for skew detection and segmentation of a document into lines. He has used a vertical profile for segmentation of lines into characters. Patel and Desai (2013) have proposed handwritten character recognition for Gujarati. They have used a tree and k-NN as a classifier. They have used structural and statistical features. They have achieved an accuracy of 63%. Desai (2015) has proposed a handwritten Gujarati alphabets identification system. He has used SVM with a polynomial kernel for classification. He has used a hybrid feature set which includes aspect ratio, extent, and divided an image into 4x4 sub-images. The author has achieved an accuracy of 86.66%. Vyas and Goswami (2015) have proposed the classification of handwritten Gujarati numeral using k-NN, SVM, and Back Propagation Neural Network. They have used spatial and transform domain features. They have the achieved highest accuracy of 93.6% using k-NN and DFT features. Kumar and Gupta (2018) have presented work on offline handwritten Gurmukhi word recognition. They have used a deep neural network as a classifier. They have used a word segmentation method. They have used LBP, directional, and regional features. They have used a mapping method to display text. They have achieved an accuracy of 99.3%.