Spatially varying reduction of haze in images

One or more tangible non-transitory computer-readable media storing computer-readable instructions that when executed by one or more processors cause the one or more processors to perform operations, the operations comprising:

• determining an image type associated with an input image;
• selecting a respective size for each of a plurality of feature windows based on the image type associated with the input image;
• generating, based on the input image, a plurality of feature vectors associated with the plurality of feature windows for a point among a plurality of points in the input image;
• providing the plurality of feature vectors as an input to one or more machine-learned models to generate a noise magnitude map associated with an amount of noise at the point;
• generating, based on the noise magnitude map and a guided filter, a refined noise magnitude map;
• generating a denoised image based on application of the refined noise magnitude map to the input image;
• and
• generating a color corrected denoised image based on performing one or more color correction operations on the denoised image including replacing low-frequency color space components of the denoised image with low-frequency color space components of the input image.

A computing system comprising:

• one or more processors;
• one or more non-transitory computer-readable media storing instructions that when executed by the one or more processors cause the one or more processors to perform operations comprising:
• determining an image type associated with an input image;
• selecting a respective size for each of a plurality of feature windows based on the image type associated with the input image;
• generating, based on the input image, a plurality of feature vectors associated with the plurality of feature windows for a point among a plurality of points in the input image;
• providing the plurality of feature vectors as an input to one or more machine-learned models to generate a noise magnitude map associated with an amount of haze noise at the point;
• generating, based on the noise magnitude map and a guided filter, a refined noise magnitude map;
• generating a denoised image based on applying the refined noise magnitude map to the input image;
• and
• generating a color corrected denoised image based on performing one or more color correction operations on the denoised image including replacing low-frequency color space components of the denoised image with low-frequency color space components of the input image.

A computer-implemented method of noise reduction, the computer-implemented method comprising:

• determining, by a computing system comprising one or more processors, an image type associated with an input image;
• selecting, by the computing system, a respective size for each of a plurality of feature windows based on the image type associated with the input image;
• generating, by the computing system, based on the input image, a plurality of feature vectors associated with the plurality of feature windows for a point among a plurality of points in the input image, wherein generating a first feature vector among the plurality of feature vectors comprises:
• determining, for a first feature window among the plurality of feature windows, a dark pixel;
• determining a hue value, a saturation value, a luminance intensity value, and a smallest channel value of the dark pixel;
• and
• generating, for the first feature window, the first feature vector which includes the hue value, the saturation value, the luminance intensity value, and the smallest channel value of the dark pixel;
• providing, by the computing system, the plurality of feature vectors as an input to one or more machine-learned models to generate a noise magnitude map associated with an amount of noise at the point;
• generating, by the computing system, based on the noise magnitude map and a guided filter, a refined noise magnitude map;
• generating, by the computing system, a denoised image based on applying the refined noise magnitude map to the input image;
• and
• generating, by the computing system, a color corrected denoised image based on performing one or more color correction operations on the denoised image.