Fast 16-bit Fixed-Point 8x8 IDCT Approximations
Lijie Liu, Trac Duy Tran, Pankaj Topiwala · 2006
In this paper, we propose two 16-bit fixed-point IDCT schemes, i.e., the rotation-based approximation and the lifting-based approximation. Both schemes can pass IEEE1180 tests and maintain satisfactory compatibility with the floating-point IDCT. The structures of these two schemes are derived from Loeffler's plane rotation-based factorization of the IDCT matrix. Coefficients in the structures are approximated carefully by dyadic rationales, leading to the fast multiplierless implementation. The drifting performance in H.263+, MPEG-2 and MPEG-4 is also demonstrated.