Design of approximated DCT with large size and its application to image matching

摘要

We present the design of approximated discrete cosine transform (DCT) with large size and its performance. The approximated DCT is designed using Wang's matrix factorization algorithms with fixed-point parameters and the size greater than 2, k > 4 is focused. So far, the specific parameters and performance of approximated DCT with large size have not been presented. In addition, we apply the sign of approximated DCT to image matching. Compared to floating-point DCT, the superior performance of the sign of the approximated DCT is demonstrated through experimental results.