A content-dependent fast DCT for low bit-rate video coding

A new content-dependent fast discrete cosine transform (DCT) algorithm is introduced. Since it requires less than half of the samples of an 8 x 8 block as input and produces only the first three DCT coefficients, it is far less complex than conventional full fast DCT algorithms. It is shown that these three coefficients are sufficient for the majority of the blocks for motion-compensated low bit-rate video coding. For other blocks, a conventional full-coefficient DCT is employed. Experiments with both low and high complexity sequences show that the content-dependent algorithm reduces the computational load of the DCT by a factor > 2x without loss in image quality and a factor 4x, if a slight PSNR loss is acceptable.