In this paper, we propose a low-cost memory reduction scheme which can redu
ce the memory requirement for full HDTV decoding from 12 MB to 4 MB. This s
cheme uses a switchable architecture that allows HDTV decoding with scalabl
e memory reduction ratios. Depending on the input format and the amount of
available memory, the scheme performs 1/4 or 1/2 memory reduction or no mem
ory reduction. The 1/2 memory reduction is achieved by performing a block-b
ased Hadamard transform followed by appropriate scalar quantization. The Ha
damard transform has good signal energy compaction and a low computational
cost. Appropriately designed nonuniform scalar quantizers take advantage of
the statistics of the Hadamard transform coefficients and compress the dat
a to match the 1/2 memory reduction target. The 1/4 memory reduction is ach
ieved by 1/2 horizontal decimation followed by the 1/2 memory reduction sch
eme. Experimental results show that the proposed memory reduction schemes a
chieve good performance at very low computational cost, which makes them Ve
ry attractive for digital TV applications.