Study On The Printed Chinese Character Font Recognition Based Wavelet Transforms

Along with the progress of science and technology, the current society has entered into the rapid-developing information age, and the workload of processing paper-based documents grows day by day. Currently the most valid method of processing such paper-based documents is to transform them to electronic documents via computer, i.e. the so called recurrence of original text. Optical character recognition (OCR) and font recognition are two key technologies to realize recurrence of original text. The OCR has been studied for a long time and its related technology has almost been mature. The font recognition, however, has not been paid enough attention and its study is still at an initial stage. Font information can provide important help for analysis, understanding and restoration of the layout of printed sheet, so this paper mainly focuses on the study of printed Chinese font recognition on the basis of the current available research achievement. The research work of this paper has done helpful effort for further improvement of font recognition technology. Some research results of this thesis have potential application foreground.A printed Chinese font recognition system is designed in this thesis. The main idea of the algorithm includes two parts. Firstly, the Chinese character consists of four basic strokes, which are horizontal, vertical, left-falling and right-falling strokes, and different Chinese font has obvious differences in the same stroke. Secondly, the Wavelet transformation has good local direction analysis ability, so it is convenient to extract the basic stroke feature. On the basis of the above analysis, a novel recognition method for standard Chinese font recognition is put forward in this thesis.The implementation procedure of the algorithm is divided into three phases. First, the Chinese character image is decomposed via wavelet transformation. Secondly, both the energy and energy ratio features of the four typical strokes are extracted. Finally, a three-layer BP artificial neural network is applied to classify the commonly-used six Chinese fonts. Experiments show that the identification method proposed in this paper can recognize the single printed Chinese font validly.