Design And Implementation Of Offline Character Recognition System Based On Android

Nowadays, with the popularization of smart mobile terminal, terminal-based applications are emerging with a large amount of requirement and applying market in character recognition research. It can be done on mobile phones to import business cards, identify road signs and translate text with the help of character recognition system, which brings large convenience to people's life. However, most of the character-recognition program in perform are based on PC, some can just recognize simple texts on mobile phones, which can't meet the requirement to import character in daily life. Therefore, an Android-based character-recognition system is designed here.Based on Android and NDK(Native Development Kit) frame, the system achieves the calls between Java upper framework and C/C++ native codes, since the latter is more efficient in image processing and pattern recognition than the former. After studying and experimenting some kinds of the preprocessing and pattern recognition algorithm, meeting the requirement of offline and complex background recognition, summarizing the key technical problems in implementing the system, the corresponding solution is given here. Innovative working in this paper is summarized as follows.(1) An Android based character-recognition system which can do recognition in complex background is designed here. All the modules are put in mobile phones to do offline recognition. Comparing to the former two develop modules, it saves development costs, brings better user experience, and uses little additional data traffic.(2) Improvement for the preprocess proposal. Existing image preprocessing algorithms are aiming at powerful processing-performance PCs. Given that the complexity of background image to be recognized and the characteristic of offline recognition in Android, here achieving the following improvement: Stretching the gray scale of image on the base of traditional Otsu to make binary effect better; Proposing a method of measuring the tilt angle through calculating the height of black pixels, which can shrink the amount of calculation and make redress effect better; Improving the projection segmentation method and doing cluster analysis on character width on the base of two-dimensional projection, which can reduce the incorrect segmentation resulting from character adhesions or the gap inside the character.(3) In the recognition phase, the two-classifiers are based on template matching. First using anti-noise performance statistical characteristics to rough classify all of the template samples and screens 50 candidate character set. Then structural features of these subdivisions candidate character set to elect the final character. Compared to the direct using template matching method, the two-classifiers not only to improve the recognition rate but also ensure the recognition rate.Owing to the contrast between the functions and performance of similar product, the Android-based character-recognition system has better real-time performance, higher recognition rate of character in complex background and can meet the requirement of importing character on mobile phone in daily life.