| 摘要 |
<p>PURPOSE:To guarantee certain constant picture quality set in advance by weighting a quantization error with visual characteristic, and separating an MSP from an LSP by using weighted quantization error. CONSTITUTION:A separation means 13 which weights the quantization error of each transformation coefficient when NXN discrete consine-transformed transform coefficients are quantized by the visual characteristic, and separates NXN quantized representative values are bi-sected to the region (MSP) including the low frequency component of a predictive error signal and the region (LSP) including the high frequency component of the predictive error signal is provided. The MSP is transmitted on a variable speed channel 14 with higher priority, and the LSP on a variable speed channel 15 with lower priority. In such a way, only the LSP not being prominent visually even by deletion is deleted when a network is set in a congested state, and the picture quality with constant level set in advance can be guaranteed by the MSP considering the visual characteristic even when the LSP is deleted.</p> |
