Open AccessCamera Captured Handwritten Kannada Character Recognition
Author(s) -
Vinod H C*,
S. K. Niranjan
Publication year - 2019
Publication title -
international journal of engineering and advanced technology
Language(s) - English
Resource type - Journals
ISSN - 2249-8958
DOI - 10.35940/ijeat.f9120.088619
Subject(s) - artificial intelligence , computer science , pattern recognition (psychology) , character (mathematics) , sobel operator , feature extraction , support vector machine , connected component labeling , centroid , feature (linguistics) , intelligent word recognition , optical character recognition , computer vision , segmentation , edge detection , image segmentation , intelligent character recognition , image processing , image (mathematics) , character recognition , mathematics , scale space segmentation , linguistics , philosophy , geometry
Optical Character Recognition (OCR) is an automatic reading of text components that are optically sensed to translate human-readable characters into machine-readable codes. In handwritten the style of writing vary from person to person, so it is very challenging task to segment and recognize the characters. In this paper we are proposing segmentation and feature extraction techniques to recognise camera captured, handwritten Kannada documents. The segmentation is done by using projection profile technique & Connected Component Analysis (CCA). The pre-processing technique to detect the edges of Kannada character, we have proposed our own technique by combining of Sobel and Canny edge detection. The feature selection and extraction is done in two level, global and local features. Global features are extracted from entire image. In local feature extraction we divided an input character image in to four quadrate based on centroid of character and we will extract local features from all quadrates rather than whole image. We have used Support vector machine (SVM) to classify the handwritten Kannada characters. To evaluate the efficiency of proposed system we have used KHDD dataset, our own document and character dataset. The experimental results shows that our proposed features selection and extraction achieved 96.31% of accuracy, results are encouraging.