2DPCA and IMLDA method of feature extraction for online handwritten Tibetan recognition

Feature extraction not only extracts the best feature which suits for the pattern classification from the original information, but also reduces the dimensions of sample in great degree. It is the significant part in the area of pattern recognition. Firstly, to extract principal component of Tibetan characters features by 2DPCA method, which can make within-class matrix no longer singularity and, the feature vectors extract from Tibetan characters have relatively high independence. And then, reducing dimension and compressing feature matrix by the IMLDA (image matrix liner discriminate analysis) approach. Finally, the reduction dimensionality Tibetan characters feature is applied to the system of online handwritten Tibetan recognition and the recognition rate increase slightly; meanwhile the paper analyzed the reasons that influence the Tibetan characters recognition.