Semantic-aware auto white balance

A computerized method comprising:

• receiving an input image;
• identifying pixels corresponding with an object in the input image;
• identifying an object type for the object in the input image as one of a plurality of object types that each has a corresponding reference color distribution, the plurality of object types including one or more skin object types and one or more other object types that each has a corresponding reference color distribution;
• selecting, from a plurality of reference color distributions for the plurality of object types, a reference color distribution based on the identified object type, the selected reference color distribution generated from a plurality of reference images of objects of the identified object type;
• determining one or more image processing settings by iteratively adjusting the image processing settings using a loss function that weights each pixel according to a probability that the pixel corresponds to the object and minimizes a difference between the selected reference color distribution and values of the pixels corresponding with the object when applying the one or more image processing settings to the input image, wherein the one or more image processing settings comprise one or more adjustable image settings of an image processing application applied to the input image that impact white balance;
• and
• generating a white balanced image by applying the one or more image processing settings to the input image, wherein the one or more image processing settings are applied to the entire input image.

A computer system comprising:

• a processor;
• and
• a computer storage medium storing computer-useable instructions that, when used by the processor, causes the computer system to perform operations comprising:
• receiving an input image;
• identifying pixels corresponding an object in the input image;
• identifying an object type for the object in the input image as one of a plurality of object types that each has a corresponding reference color distribution, the plurality of object types including one or more skin object types and one or more other object types that each has a corresponding reference color distribution;
• selecting, from a plurality of reference color distributions for the plurality of object types, a reference color distribution based on the identified object type, the selected reference color distribution generated from a plurality of reference images of objects of the identified object type;
• determining one or more image processing settings by iteratively adjusting the image processing settings using a loss function that weights each pixel according to a probability that the pixel corresponds to the object and minimizes a difference between the selected reference color distribution and values of the pixels corresponding with object when applying the one or more image processing settings to the input image, wherein the one or more image processing settings comprise one or more adjustable image settings of an image processing application applied to the input image that impact white balance;
• and
• generating a white balanced image by applying the one or more image processing settings to the input image, wherein the one or more image processing settings are applied to the entire input image.