In this paper, we have proposed an Adaptive Radial Basis Function Network (ARBFN) based writer adaptation for online handwriting recognition. We used RBF Network for both training (writer independent recognition system) and adaptation (writer dependent recognition system) using incremental learn¬ing. The adaptation is performed by either adding a center to the present network or modifying the weights of the existing proto¬type vectors using standard LMS gradient descent, whenever a misclassification is reported. ARBFN is evaluated on devanagari characters and numerals, collected in house.
Online handwriting recognition research has recently received significant thrust. Specifically for Indian scripts, handwriting recognition has not been focused much till in the near past. However, due to generous Government funding through the group on Technology Development for Indian Languages (TDIL) of the Ministry of Communication & Information Technology (MC&IT), Govt. of India, research in this area has received due attention and several groups are now engaged in research and development works for online handwriting recognition in different Indian scripts. An extensive bottleneck of the desired progress in this area is the difficulty of collection of large sample databases of online handwriting in various scripts. Towards the same, recently a user-friendly tool on Android platform has been developed to collect data on handheld devices.
In this paper, we present our recent study of a data driven approach to combining multiple SVM classifiers with RBF kernels each being trained with a distinct feature vector. The SVM classifiers in our ensemble are ranked based on their increasing order of average performance on the validation sample sets. The outputs of the SVM classifiers are combined based on a weighted average strategy which uses the above ranks of the underlying SVMs to determine the respective weights. In the present study, we design four sets of different feature vectors representing online handwritten words. Simple enation of these feature vectors does not help much in improving the recognition accuracy compared to the best performing feature vector among the four.
Added on September 5, 2017
Contributed by : OHWR Consortium
Product Type : Research Paper
License Type :
System Requirement :
Author : A. Srimany ,S. Dutta Chowdhuri,U. Bhattacharya,S. K. Parui
A reasonably large database of online handwritten Bangla characters has been developed. Such a handwritten character sample is composed of one or more strokes. Seventy five such stroke classes have been identified on the basis of the varying handwriting styles present in the character database. Each character sample is a sequence of strokes emanating from these stroke classes. Another database of handwritten Bangla strokes has been developed from the character database. This is the first such database for Bangla script. Certain stroke level features are defined on the basis of certain extremum points which represent the stroke shape reasonably well.
This work describes the development of Assamese online numeral recognition system using Hidden Markov Models (HMM) and Support Vector Machines (SVM). Preprocessed (x,y) coordinates and their first and second derivatives at each point are used as features for both the modeling techniques. The two systems are developed individually using HMM and SVM. The results from both the systems are then combined using two different approaches. In the first approach, the scores from both the classifiers are directly merged and an improvement in performance is observed in the combined system (comb−1). In the second approach,the confusion patterns from HMM and SVM classifiers are also analyzed. Based on this, the results are further combined to obtain a final hybrid numeral recognizer with anenhanced performance (comb−2). The HMM, SVM, Comb-1and Comb-2 systems provide average recognition performance of 96.5, 96.8, 98 and 98.3, respectively.