Muhammad Faisal Zafar

Recognition of Online Isolated Handwritten Characters by Backpropagation Neural Nets Using Sub-Character Primitive Features

In online handwriting recognition, existing challenges are to cope with problems of various writing fashions, variable size for the same character, different stroke orders for the same letter, and efficient data presentation to the classifier. The similarities of distinct character shapes and the ambiguous writing further complicate the dilemma. A solitary solution of all these problems lies in the intelligent and appropriate extraction of features from the character at the time of writing. A typical handwriting recognition system focuses on only a subset of these problems. The goal of fully unconstrained handwriting recognition still remains a challenge due to the amount of variations found in characters. The handwriting recognition problem can be considered for various alphabets and at various levels of abstraction. The main goal of the work presented in this paper has been the development of an on-line handwriting recognition system which is able to recognize handwritten characters of several different writing styles. Due to the temporal nature of online data, this work has possible application to the domain of speech recognition as well. The work in this research aimed to investigate various features of handwritten letters, their use and discriminative power, and to find reliable feature extraction methods, in order to recognize them. A 22 feature set of sub-character primitive features has been proposed using a quite simple approach of feature extraction. This approach has succeeded in having robust pattern recognition features, while maintaining features domain space to a small, optimum quantity. Backpropagation neural network (BPN) technique has been used as classifier and recognition rate up to 87% has been achieved even for highly distorted handwritten characters

Neural Nets for On-line Isolated Handwritten Character Recognition: A Comparative Study

Handwriting processing is a domain in great expansion which begins to see several industrial realizations. The field of personal computing has begun to make a transition from the desktop to handheld devices, thereby requiring input paradigms that are more suited for single hand entry than a keyboard. Online handwriting recognition allows for such input modalities. Handwriting recognition has always been a tough problem because of the handwriting variability, ambiguity and illegibility. This paper describes a simple approach involved in online handwriting recognition. Conventionally, the data obtained needs a lot of preprocessing including filtering, smoothing, slant removing and size normalization before recognition process. Instead of doing such lengthy preprocessing, this paper presents a simple approach to extract the useful character information. The whole process requires no preprocessing and size normalization. The method is applicable for off-line character recognition as well. This is a writer-independent system based on two neural net (NN) techniques: back propagation neural network (BPN) and counter propagation neural network (CPN). Performances of BPN and CPN are tested for upper-case English alphabets for a number of different styles from different peoples

Performance analysis of a proposed smoothing algorithm for isolated handwritten characters

This paper describes an online isolated character recognition system using advanced techniques of pattern smoothing and Direction Feature (DF) extraction. The composition of direction elements and their smoothing are directly performed on online trajectory, and therefore, are computationally efficient. We compare recognition performance when DFs are formulated using Smoothed Direction Vectors (SDV) and Unsmoothed Direction Vectors (UDV). In experiments, direction features from original pattern yielded inferior performance, whereas primitive sub-character direction features using smoothed direction-encoded vectors made significant difference. Recognition rates were improved by about 7% and 5% using SDV when compared with UDV and smoothed with Moving Average (MA) technique, respectively.