![]() | Only 14 pages are availabe for public view |
Abstract Advancements in technologies in the recent ten years brought to us the prorrusmg technology of optical character recognition (OCR). There is a big problem when identifying text with different fonts because of the difference between the same letter shapes in different fonts. The previous research is concentrated for systems with similar characteristics fonts. So, in our research, we concern with the different characteristics multi-font Arabic machine- printed optical text recognition. This is done by identifying the font type before recognizing the characters which significantly increases the total mean recognition rate of the proposed system. There are three main contributions in our research. The first contribution is in the cursive Arabic character segmentation. This was done by designing a constant amplitude (low variations) passing filter. This filter operates on the output of the vertical axis profile of the word. The filter output is a locus of the characters separations. With the aid of this filter, the correction ratio of this algorithm reaches 98%. The second contribution is in the field of the character recognition using scale invariant detectors. This part uses eight techniques to get the best one which is suitable for Arabic OCR. A complete study was performed for all the eight scale invariant detectors and approved that the Harris Laplace is the best one for describing the Arabic characters with the aid of the gradient descriptor and the k-rneans clustering. The last contribution is in the field of multi fonts Arabic OCR. The proposed optical Arabic font recognition (AOFR) algorithm suggested achieved a correction ratio of 100% over 15 fonts. This algorithm enhances the mean recognition rate of the character recognition for multi-fonts from 40% to 99%. |