Fig. 3From: Enlarging smaller images before inputting into convolutional neural network: zero-padding vs. interpolationSynaptic weights (a, b, c, d, e, f, g, h, and i) are the same for all convolution windows in CNNBack to article page