Feature extraction of label character based on projection transformation

A method of label character image feature extraction based on projection transformation, wavelet analysis and discrete cosine transform is proposed. First, it transforms the Cartesian coordinate into polar coordinates using the centroid as the pole, and makes circular projection calculation in order to transform 2-D label characters into 1-D projection curves. Then, it gets the approximate signals and the details signals of the projection curves using wavelet decomposition. Finally, it separately does discrete cosines transform for these signals in order to get eigenvector of the label character. Experiments show that this method has the scale invariance, rotation invariance, good robustness and real-time.