Encoder, decoder, encoding method, and decoding method

1. An encoder, comprising:

processing circuitry; and

memory, wherein

using the memory, the processing circuitry:

generates a predicted image of an input image that is a current image to be encoded, based on generated data output from a generator network in response to a reference image being input to the generator network, the generator network being a neural network;

calculates a prediction error by subtracting the predicted image from the input image;

generates an encoded image by at least transforming the prediction error;

feeds back, to the generator network, a probability that the predicted image matches the input image by inputting the input image and the predicted image to a discriminator network, the discriminator network being a neural network and constituting a generative adversarial network (GAN) with the generator network; and

updates the generator network and the discriminator network to reduce difference between the input image and the predicted image and increase accuracy of discriminating between the input image and the predicted image,

wherein the reference image is a processed image included in a picture, the picture including the input image, and

wherein in generating the predicted image, the processing circuitry:

generates a first intra-predicted image by a first intra prediction based on the reference image and an intra prediction parameter, the intra prediction parameter being obtained as the generated data output from the generator network in response to the reference image being input to the generator network;

generates a second intra-predicted image of the input image by a second intra prediction based on the reference image;

selects, as the predicted image, from among the first intra-predicted image and the second intra-predicted image; and

when the processing circuitry selects the second intra-predicted image, calculates the prediction error by subtracting the second intra-predicted image from the input image in calculating the prediction error.

2. The encoder according to claim 1 , wherein

the generator network is a hierarchical network that includes an input layer, a hidden layer, and an output layer.

3. A decoder, comprising:

processing circuitry; and

memory, wherein

using the memory, the processing circuitry:

generates a decoding prediction error by performing at least inverse transform on an encoded image that is a current image to be decoded;

generates a predicted image of the encoded image, based on generated data output from a generator network in response to a reference image being input to the generator network, the generator network being a neural network;

generates a decoded image by adding the decoding prediction error to the predicted image;

feeds back, to the generator network, a probability that the predicted image matches the decoded image by inputting the decoded image and the predicted image to a discriminator network, the discriminator network being a neural network and constituting a generative adversarial network (GAN) with the generator network; and

updates the generator network and the discriminator network to reduce difference between the decoded image and the predicted image and increase accuracy of discriminating between the decoded image and the predicted image,

wherein the reference image is a processed image included in a picture, the picture including the encoded image, and

wherein in generating the predicted image, the processing circuitry:

generates a first intra-predicted image by a first intra prediction based on the reference image and an intra prediction parameter, the intra prediction parameter being obtained as the generated data output from the generator network in response to the reference image being input to the generator network;

generates a second intra-predicted image of the encoded image by a second intra prediction based on the reference image;

selects, as the predicted image, from among the first intra-predicted image and the second intra-predicted image; and

when the processing circuitry selects the second intra-predicted image, generates the decoded image by adding the decoding prediction error to the second intra-predicted image in generating the decoded image.

4. The decoder according to claim 3 , wherein

the generator network is a hierarchical network that includes an input layer, a hidden layer, and an output layer.

5. An encoding method, comprising:

generating a predicted image of an input image that is a current image to be encoded, based on generated data output from a generator network in response to a reference image being input to the generator network, the generator network being a neural network;

calculating a prediction error by subtracting the predicted image from the input image;

generating an encoded image by at least transforming the prediction error;

feeding back, to the generator network, a probability that the predicted image matches the input image by inputting the input image and the predicted image to a discriminator network, the discriminator network being a neural network and constituting a generative adversarial network (GAN) with the generator network; and

updating the generator network and the discriminator network to reduce difference between the input image and the predicted image and increase accuracy of discriminating between the input image and the predicted image,

wherein the reference image is a processed image included in a picture, the picture including the input image, and

wherein the generating the predicted image includes:

generating a first intra-predicted image by a first intra prediction based on the reference image and an intra prediction parameter, the intra prediction parameter being obtained as the generated data output from the generator network in response to the reference image being input to the generator network;

generating a second intra-predicted image of the input image by a second intra prediction based on the reference image;

selecting, as the predicted image, from among the first intra-predicted image and the second intra-predicted image; and

when the second intra-predicted image is selected as the predicted image, calculating the prediction error by subtracting the second intra-predicted image from the input image.

6. A decoding method, comprising:

generating a decoding prediction error by performing at least inverse transform on an encoded image that is a current image to be decoded;

generating a predicted image of the encoded image, based on generated data output from a generator network in response to a reference image being input to the generator network, the generator network being a neural network;

generating a decoded image by adding the decoding prediction error to the predicted image;

feeding back, to the generator network, a probability that the predicted image matches the decoded image by inputting the decoded image and the predicted image to a discriminator network, the discriminator network being a neural network and constituting a generative adversarial network (GAN) with the generator network; and

updating the generator network and the discriminator network to reduce difference between the decoded image and the predicted image and increase accuracy of discriminating between the decoded image and the predicted image,

wherein the reference image is a processed image included in a picture, the picture including the encoded image, and

wherein the generating the predicted image includes:

generating a first intra-predicted image by a first intra prediction based on the reference image and an intra prediction parameter, the intra prediction parameter being obtained as the generated data output from the generator network in response to the reference image being input to the generator network;

generating a second intra-predicted image of the encoded image by a second intra prediction based on the reference image;

selecting, as the predicted image, from among the first intra-predicted image and the second intra-predicted image; and

when the second intra-predicted image is selected as the predicted image, generating the decoded image by adding the decoding prediction error to the second intra-predicted image.