Recognition of offline handwritten Devanagari characters using new mask-based approach, histogram of oriented gradients and AdaBoost

The world is growing with the new technologies for machine learning. Day by day there are incredible developments in the area of pattern recognition for the applications such as institutional record keeping, ancient documents preservation, postal address sorting, signature verification, etc. Handwritten character recognition is one such application grabbing lot of attention for the machines to learn. Handwritten Devanagari script is used in the proposed work which is very difficult to understand by the machines due to many similar shapes incorporated with it. So, pre-processing before extracting the features from the character plays very important role in increasing the accuracy. By keeping this in mind, a new framework for handwritten Devanagari character recognition is proposed where entropy -based skew correction is used to correct the skew of the characters and Mask-based approach is used which efficiently removes the header line and returns the header free character. The histograms of oriented gradients (HOG) features are extracted from the header free characters and provided for non-parametric dimensionality reduction. These features are further classified using AdaBoost ensemble boosting method and achieved a very good recognition accuracy of 98.43% and 98.68% on V2DMDCHAR and ISIDCHAR databases respectively.

Automated summary

The world is witnessing incredible developments in the field of machine learning for pattern recognition in various applications. Handwritten character recognition, especially in Devanagari script, poses challenges due to similar shapes. Pre-processing techniques like entropy-based skew correction and Mask-based approach are used to increase accuracy. The extracted Histograms of Oriented Gradients (HOG) features are further classified using AdaBoost ensemble boosting method, achieving high recognition accuracy.