Image filtering apparatus, image decoding apparatus, and image coding apparatus
A filtered image with enhanced image quality is output. Provided is a loop filter including: a filtering processing unit configured to apply a filter to an unfiltered image; and an ALF unit configured to receive input of a DL image to which a filter is applied by a first filtering processing unit out of the filtering processing unit. The ALF unit further receives input of the unfiltered image.
1. An image filtering apparatus for filtering an input image, the image filtering apparatus comprising:
a loop filter that includes (i) at least a first neural network filter and a second neural network filter, and (ii) an adaptive loop filter,
wherein:
the first neural network filter receives (i) the input image and (ii) an image of a quantization parameter which is defined by an array of two dimensions, and outputs a first output image,
the second neural network filter receives the input image or the first output image, and outputs a second output image, and
the adaptive loop filter receives (i) the input image, (ii) the first output image, and (iii) the second output image, and outputs a filtered image, as a third output image.
2. The image filtering apparatus according to claim 1 , wherein:
a neural network includes a plurality of layers in each of which filters applied, and
the adaptive loop filter receives (i) the first output image in which the filters are applied in all of the plurality of layers and (ii) the second output image in which the filters are applied in one or multiple layers of the plurality of layers.
3. An image decoding apparatus for decoding an image, the image decoding apparatus comprising:
the image filtering apparatus according to claim 1 as a filter to be applied to a decoded image.
4. An image encoding apparatus for encoding an image, the image encoding apparatus comprising:
the image filtering apparatus according to claim 1 as a filter to be applied to the input image.
5. The image filtering apparatus according to claim 1 , wherein:
the second neural network filter is a subset of the first neural network filter, and
the second neural network filter outputs one or multiple inter-mediate layers of the first neural network filter as the second output image.