| 摘要 |
<p>PROBLEM TO BE SOLVED: To provide a device and method for decoding by which the quality of a reproduced image can be improved without increasing the amount of information by reproducing the information lost by quantization through inference at the time of decoding data which are compression-encoded by DCT conversion. SOLUTION: When the data compressed by DCT conversion are decoded, the high-accuracy coefficient data of a high-accuracy frequency region is generated through inference with respect to the coefficient data of a frequency region outputted from a reverse quantization circuit 12 by means of an information inference adding circuit 13 provided in the preceding stage of a reverse DCT conversion circuit 14. For the inference, a neural network model is used. By reversely converting the coefficient data of the high-accuracy frequency region estimated from the reversely quantizing means through inference, high-accuracy data can be decoded.</p> |
