Apparatus and method for performing artificial intelligence encoding and decoding on image by using low-complexity neural network

1. A server for providing an image using artificial intelligence (AI), the server comprising:

a processor configured to execute one or more instructions stored in the server to:

obtain, as a reduction feature map having a resolution lower than a resolution of an original image, a residual image between a first reduced image and a second reduced image that are downscaled from the original image;

obtain a first image by inputting, to a downscale neural network (NN), the first reduced image and the reduction feature map, and by adding an output image of a last layer of the downscale NN and the second reduced image;

encode the first image to generate image data; and

provide an electronic device with the image data.

2. The server of claim 1 , wherein the processor is further configured to:

obtain a plurality of first reduced images comprising pixels located at different points from each other within pixel groups of the original image; and

obtain, as the reduction feature map, a plurality of residual images between the plurality of first reduced images and the second reduced image.

3. The server of claim 2 , wherein a sum of a number of the plurality of first reduced images and a number of the plurality of residual images is equal to a number of input channels of a first layer of the downscale NN.

4. The server of claim 1 , wherein the processor is further configured to obtain an edge map corresponding to the original image as the reduction feature map.

5. The server of claim 1 , wherein the processor is further configured to:

upscale the second reduced image is downscaled from the original image; and

obtain, as the reduction feature map, the residual image between the second reduced image that is upscaled after being downscaled and the first reduced image.

6. The server of claim 1 , wherein output data of any one layer of a plurality of layers of the downscale NN is added to output data of preceding layers prior to the any one layer, and a sum of the output data of the any one layer and the output data of the preceding layers is input to a next layer of the any one layer.

7. A method for providing an image by a server configured to use artificial intelligence (AI), the method comprising:

obtaining, as a reduction feature map having a resolution lower than a resolution of an original image, a residual image between a first reduced image and a second reduced image that are downscaled from the original image

obtaining a first image by inputting, to a downscale neural network (NN), the first reduced image and the reduction feature map and by adding an output image of a last layer of the downscale NN and the second reduced image;

encoding the first image to generate image data; and

providing an electronic device with the image data.

8. The method of claim 1 , further comprising:

obtaining a plurality of first reduced images comprising pixels located at different points from each other within pixel groups of the original image; and

obtaining, as the reduction feature map, a plurality of residual images between the plurality of first reduced images and the second reduced image.

9. The method of claim 8 , wherein a sum of a number of the plurality of first reduced images and a number of the plurality of residual images is equal to a number of input channels of a first layer of the downscale NN.

10. A computer-readable recording medium having recorded thereon a program for executing the method of claim 7 .