Regional image correspondence matching method for SIMD processing

We present a block based image matching framework in order to estimate motion field and for matching pre-learned face templates to grayscale input images. The proposed algorithm is designed particularly for SIMD type massively parallel processing capable of binary operations and global result read-out. Face detection is performed by facial templates of different sizes, which are slid through the image in order to detect frontal faces with low resolution (down to 20 times 20 pixels). The applications illustrate the usefulness of the proposed non-parametric image matching scheme.