A study on the use of Gabor features for Chinese OCR

Abstract

The authors revisit the topic of Gabor feature extraction for Chinese OCR. We adopt a very simple discriminant function to construct a maximum discriminant function based character recognizer. We experiment with a simple way of forming a feature vector for each character image by extracting Gabor features using one wavelength at locations uniformly sampled with one spatial resolution. Extensive experiments on large vocabulary Chinese OCR for both machine-printed and handwritten characters are performed by using a large amount of training and testing data to demonstrate the effectiveness of the Gabor features for Chinese OCR. Using Gabor features as raw features, we have constructed several state-of-the-art Chinese OCR engines.