This paper is to realize the optical character recognition on grey scale level by adopting learning subspace method of principal component analysis(PCALSM). Compared with Arabic number images

the resolution of Chinese character images is small

which creates great difficulty in extracting the character features. And it will get worse especially when the quality of image is low. PCALSM can overcome the main shortages of classification on binary images

and keeps integrity features of character information dramatically. On the basis of PCA subspaces

training of each subspace is rotated in different ways of the supervised feedback learning algorithm; and better classification is therefore obtained. The time consuming subspace training can be accepted especially when the number of character classes is not large. Our experimental results have proved that recognition of car license plate characters (a limited Chinese character set) has been improved by PCALSM

which makes it highly worth applying this optical character recognition (OCR) method.