| 摘要 |
PURPOSE:To realize the reduction of a circuit scale by inputting input image data twice into a neural network as the same smoothing means in an image forming device such as a printer, etc. CONSTITUTION:The smoothing means 10 is constituted by the neural network which outputs a virtual correcting information such as only the correcting information of a dot size to input image data at the time of first operation and outputs the correcting information of a final dot size and position to the input of the piece of virtual correcting information at the time of second operation. In accordance with the input of picture element data within a rectangular window corresponding to respective five numbers of dots on three lines, for example, in inputted dot image data, the neural network constituting the smoothing means 10 finally outputs correcting data of the size/position of the central dot of the window. Then, the piece number of the units of an intermediate layer can be reduced so that the processing speed of the neural network can be improved. |
