Paper Details

Automatichandwritten characters recognition is a problem, which is currently gathering a lot of attention. Theability of an efficient processing small handwriting samples, such as those found on cheques and envelopes, is one of thesignificant driving forces behind this current research. This paper describes a geometry based technique for featureextraction which applies to the segmentation-based word recognition systems.In this methodology, an artificial neuralnetwork is trained to identify resemblance and patterns among different handwriting character dataset training samples anduser-entered characters. The proposed system extracts the geometric features of character and thereby, forming acharacterskeleton.The system generates feature vectors as outputs which are used to train a pattern recognition engine basedon Neural Networks which makes the system benchmarked. We acquired an accuracy of 95.2% working on a set of 108features. The Feature-Extraction methods described in this paper have performed well in classification when fed to theneural network, and pre-processing of the image using edge detection method and normalization technique are the idealchoice for degraded noisy images.

Dishant Khanna, Narina Thakur, “Feature Extraction Technique for Handwritten Character Recognition using Geometric-Based Artificial Neural Network”, International Journal of Advance Engineering and Research Development (IJAERD), Vol. 3, Issue 8, pp. 121-130, August 2016.