A new research method for Uyghur character recognition is proposed.Firstly

a character sample corpus is built and all characters and images in it are normalized.Then

correlation detection on both horizontal and vertical directions projections of test images and sample images are carried out

and then extract stroke number feature of sample images which have higher correlation with double projection of test image to get preclassification result.Finally

detect SIFT key points of test image and pre-classification results

generate orientation descriptors and registrate and pre-classification results with the highest matching number points are regarded as recognition result and output result notation Uyghur character.Experimental results show that this method can reduce number of characters sample and effectively solve difficulty caused by difference of scale and geometric distortion of test image.