Offline Handwritten Devanagari Word Recognition: An HMM Based Approach

Abstract

A hidden Markov model (HMM) for recognition of handwritten Devanagari words is proposed. The HMM has the property that its states are not defined a priori, but are determined automatically based on a database of handwritten word images. A handwritten word is assumed to be a string of several stroke primitives. These are in fact the states of the proposed HMM and are found using certain mixture distributions. One HMM is constructed for each word. To classify an unknown word image, its class conditional probability for each HMM is computed. The classification scheme has been tested on a small handwritten Devanagari word database developed recently. The classification accuracy is 87.71% and 82.89% for training and test sets respectively.

Handwritten Hindi character recognition: a review

“…Although there are pitfalls of using smoothing filters, but their impact also depends upon the features used or how they have been applied. In [24], the character images are binarised using Otsus thresholding method [25] and then, smoothed using median filters because in this work they are extracting long horizontal and vertical strokes and neglecting small curves of character. Garg et al [26] segmented touching modifiers and consonants using vertical projection profile and structural properties of the text.…”

“…Parui and Shaw et al [24] proposed word recognition technique based on horizontal and vertical strokes extraction using east and south neighbours. Scaling invariant features shape, size, and positions are determined from digital strokes and fed into hidden Markov model (HMM) classifier.…”

Real Time Handwritten Marathi Numerals Recognition Using Neural Network

“…Here, we choose the marker image is f m to be "0" everywhere except on the image border, where it is set to "1" [6,[10][11]. Figure 4 show the above preprocessing steps followed on online data.…”

“…In this work, mu ltilayer feedforward back propagation neural network has been implemented. The neural network had three layers: an input layer consisting of 35 nodes, a hidden layer consisting of 20 nodes, and an output layer 10 nodes one for each numerals [11][12][13][14][15][16] . Fig.…”

Parallel Implementation of Devanagari Text Line and Word Segmentation Approach on GPU

“…Research on Devanagari [1] character [2][3][4][5][6][7][8][9][10] and word [11][12][13] recognition is very difficult due to its challenging properties. This area of research is still open for further research due to the extent of variation among writing styles, speed, thickness of character and direction of different writers.…”