CS388, May 2020, Earlham CollegeKate Nguyenand the synthetic classifier,which is twice as fast as the former,obtained an accuracy of 92.4% over the dataset of printed Chinesecharacters.While some methods achieved significantly higher rates of accu-racy, it is worth noting that these studies used different types ofdatasets. The method that Yin et al. obtained an accuracy of 99.38%.However, the proposed targeted Chinese uppercase characters ex-clusively, which is a particularly small set of Chinese characters[14].Similarly,the methods that Zheng et al. [15] and Kim et al.[2] proposed, which achieved the accuracy rates of 98% and 92.4%,respectively, both tested the models on printed Chinese characters,which are clearer and less complicated to be recognized comparedto handwritten characters. Hence, the rates of accuracy achieved bythe methods mentioned in this section might not reflect the state-of-the-art performance of handwritten Chinese character recognition.2.2Post-processing with Statistical TextModelsA different approach to improve the accuracy of the OCR is workingon error correction in the post-processing phase instead of directlyimproving the performance of the recognition models. Most studiesin this category applied statistical text models to correct errors ofthe OCR after the recognition phase.Tseng proposed an error correction technique for OCR that iden-tifies keywords and confusable terms [11]. The algorithm first ex-tracts maximally repeated patterns from each document; then, itwill collect terms with three or over three characters. If two termsonly differ in one character, the two characters that differentiatethese two terms will be reported as a confusing pair. The authorclaimed that this post-processing method achieved an accuracyof 84% and,therefore,can be used to improve the performanceof the OCR.Liu and Cao also demonstrated a similar approachcalled a seed-based method [5]. This method consists of two stages:forming a seed of confusion set for each character and creating agraph of confusion sets, then developing a self- extension methodto extend the graph. The proposed method resulted in an increaseof confusing pairs by 36.4%. Although the impact of this study onthe performance of the OCR was unknown as this work had notbeen directly applied to the OCR model, a significant increase ofconfusing pairs might lead to a substantial improvement in theaccuracy of the OCR.Wang and Liu proposed a method that helps develop a more accu-rate OCR system for Chinese text in rare books [13]. The proposedmethod consists of N-gram, long short-term memory (LSTM), andard and forward N-gram (BF N-gram) models. Through their exper-iments, they concluded that N-gram, LSTM, or BF N-gram are notsufficient to reduce the errors of OCR. Adding rules of OCR Top 5and the probability of Top 1, where Top 5 are the candidate wordswith the highest classification probability, and Top 1 is the result ofit, provides more information to make choices.Adopting the same approach, Li et al. presented a contextual post-processing method that combines character-based bigram post-processing and word-based bigram post-processing for Chinesecharacters recognition [3]. They first executed character-based bi-gram post-processing using a forward-backward search on a largecandidate set in order to improve the accuracy and efficiency of theset. The word-based bigram post-processing was then executed ona small candidate set to improve accuracy further. They concludedthat their new post-processing method both improves accuracy andboosts the processing speed, which was 100 times faster comparedto word-based bigram post-processing.3DESIGNThis research project will study different combinations of the exist-ing methods that aim to solve the same problems, which includemethods in the recognition phase and the post-processing phase,and add an image pre-processing phase that happens before imple-menting each combination. A novel statistical text model that isapplied in the post-processing stage is proposed. The new combi-nation that consists of the novel model will be compared with thecombinations of existing methods in terms of recognition accuracy.3.1FrameworkThe handwritten Chinese characters dataset will be arbitrarily di-vided into two sets: train dataset and test dataset. The images inboth datasets will be treated using image pre-processing techniquesincluding binarization,skew correction,and denoise in order toimprove the quality of the images. This facilitates the recognitionmodel to achieve a better result. The treated train dataset will be fedto the CNN to extract the features and train the classification model.After the training stage, the treated test dataset will be fed to thetrained CNN model to classify the characters. In the post-processingphase, the errors existing in the results from the recognition phasewill be corrected to increase the accuracy rate of the OCR. The post-processing process is based on statistical text models. The modelsthat will be applied in this research include Bayesian, backwardand forward n-gram [13], and a novel model. The framework ofthis project is shown in Figure 1.Figure 1: Framework of the project3.2Pre-processing Techniques for OCR3.2.1Binarization.Binarization is the process that converts col-ored image into an image that consists of only black and whitepixels [1].Pixel values in an image will be separated in to two

+13062052269

info@desklib.com

Pre-processing Techniques in OCR

Contribute Materials

Related Documents