| 摘要 |
PURPOSE:To suppress an error in orthogonal transformation by controlling the device so that a round-off bit number being the result of calculation is decreased within a range of a limit of an input bit in addition when a block size is small. CONSTITUTION:A picture stored in a picture memory 1 is segmented into a block in response to the block size for orthogonal transformation with a block readout section 3 to form an input data block. Moreover, a coefficient register 4 expresses a base vector of orthogonal transformation in a way of a discrete signal and stores the result as a coefficient and it is fetched in a multiplication section 5 together with the input data block generated by the readout section 3. Then the multiplication section 5 multiplies the fetched input data block with the content of a coefficient memory 4 and the result is fetched in a round- off section 6. Then the round-off section 6 controls itself and the block readout section 3 in response to the block size of orthogonal transformation. Thus, when the orthogonal conversion block size is small, it is possible to eliminate waste of the carry bit of an adder section to improve the arithmetic accuracy of the multiplication section. |
