In this research, I propose a two-variable activation function “Yamatani” that satisfies the first-degree homogeneity, and realize a super-resolution convolutional neural network that is independent of the dynamic range and symmetrical about the luminance inversion.