Methods and apparatus to reduce compression artifacts in images

1. A method comprising:

separating at least a portion of an image into a first component and a second component;

reducing a first artifact in the first component to form a first cleaned component by applying a convolutional neural network to the first component;

reducing, using the first cleaned component, a second artifact in the second component to form a second cleaned component by applying a filter having coefficients based on the first cleaned component; and

combining the first cleaned component and the second cleaned component to form a cleaned image.

2. The method of claim 1 , further comprising training the convolutional neural network using luma channels of a plurality of images.

3. The method of claim 2 , wherein the convolutional neural network is trained using a cropped luma patch to recover a center pixel.

4. The method of claim 1 , wherein the image is a frame of a video signal.

5. The method of claim 1 , wherein the image is an RGB image, the first component represents a luma component, and the second component represents a chroma component.

6. The method of claim 1 , wherein the filter computes a sum of products of the second component and the coefficients.

7. An apparatus comprising:

a splitter to separate at least a portion of an image into a first component and a second component;

a convolutional neural network to reduce a first artifact in the first component to form a first cleaned component;

a filter having coefficients determined based on the first cleaned component to filter a second artifact in the second component to form a second cleaned component; and

a combiner to combine the first cleaned component and the second cleaned component to form a cleaned image.

8. The apparatus of claim 7 , wherein the convolutional network comprises three convolutional layers.

9. The apparatus of claim 7 , wherein the convolutional neural network is trained using luma channels of a plurality of images.

10. The apparatus of claim 9 , wherein the convolutional neural network is trained using a luma patch to recover a center pixel.

11. The apparatus of claim 7 , wherein the image is a frame of a video signal.

12. The apparatus of claim 7 , wherein the image is an RGB image, the first component represents a luma component, and the second component represents a chroma component.

13. The apparatus of claim 7 , wherein the filter computes a sum of products of the second component and the coefficients.

14. A non-transitory machine-readable media storing machine-readable instructions that, when executed, cause a machine to at least:

separate at least a portion of an RGB image into a luma component and a chroma component;

reduce a first artifact in the luma component to form a cleaned luma component by applying a convolutional neural network to the luma component;

reduce, using the cleaned luma component, a second artifact in the chroma component to form a cleaned chroma component by applying a filter having coefficients based on the cleaned luma component; and

combine the cleaned luma component and the cleaned chroma component to form a cleaned RGB image.

15. The non-transitory machine-readable media storing machine-readable instructions of claim 14 that, when executed, cause the machine to train the convolutional neural network using a luma patch to recover a center pixel.

16. The non-transitory machine-readable media storing machine-readable instructions of claim 14 that, when executed, cause the machine to apply the filter by computing a sum of products of the chroma component and the coefficients.