A New Contour Based Invariant Feature Extraction Approach for the Recognition of Multi-lingual Documents

Abstract

Now a day, developing a single OCR system for recognizing multi-lingual documents becomes essential to enhance the ability and performance of the existing document analysis system. Hence in this paper, we present a new technique based on contour detection and distance measure for recognizing multi-lingual characters comprising south Indian languages (Kannada, Tamil, Telugu, Malayalam, English Upper case, English Lower case, English Numerals and Persian Alphanumeric). Proposed method finds boundary for a character using contour detection and the result of contour detection is given to feature extraction scheme to obtain distinct and invariant features for identifying different characters of different languages. The method extracts invariant features by computing distance between the centroid and the pixels of contour of character image. We compare the experimental results of proposed method with result of existing methods to evaluate the performance of the method. Based on experimental results it is realized that the proposed method gives 100% accuracy with minimum expense and time. In addition, the method is invariant to Rotation, Scaling and Translation transformations (RST).