A proposed hardware reference model for spatial transformation and quantization in H.264

This paper presents three Very Large Scale Integration prototypes to exploit spatial redundancy in the H.264 standard. The proposed architectures are: (1) forward 4 × 4 integer approximation of DCT transform and quantization, which is applied to all blocks of a frame, (2) the 4 × 4 Hadamard transform and quantization that is applied to the DC coefficients of the luma component when the macroblock is encoded in 16 × 16 intra prediction mode, and (3) the 2 × 2 Hadamard transform and quantization that is applied to the DC coefficients of the chroma component as a second level in the transformation hierarchy. The developed algorithms are adopted by the H.264 standard. A performance analysis shows that the architectures satisfy the real-time constraints required by different digital video applications.