A Feature Extraction Algorithm of Affine Invariant based on Region Partition

Extracting features is one of the core issues in pattern recognition, in which affine invariant is important method to recognize object under complex environment. To solve the issues of the error increasing and the low efficiency for affine invariant extraction algorithms, a improved feature extraction algorithm based on region partition is proposed. First, the algorithm gets the centroid and optimalvertex of the binarized object image. Then, a new partition strategy of the affine region is applied. Finally, the affine invariant vecter will be obtained by the area ratio of the affine region. Experiments on Columbia University’s Fish Database show that the invariant extracted by the proposed algorithm satisfies affine transformation properties. Compared with three popular affine invariant extraction algorithms, the proposed algorithm can achieve a better performance. The extracted invariant has a well ability to distinguish objects.