An acquisition condition is decided for a second image of improved quality. Values xi resulting from down sampling brightness values of an input first image are accepted by an input layer. A filter is scanned and a convolutional computation performed in a convolutional layer. Outputs z1 to z4 of the convolutional layer and a first image acquisition condition v=(v1, v2, v3, v4) of the first image are accepted by an output layer and a second acquisition condition y is computed by the output layer.