An image processing apparatus includes a main unit configured to convert an input image into information of a first feature amount using a first convolutional neural network having at least two layers, an input unit configured to convert the input image into information of a second feature amount using a second convolutional neural network, and an output unit configured to convert information of a third feature amount generated by adding the information of the first feature amount and the information of the second feature amount to each other, into an output image using a third convolutional neural network.