Identifying and modifying cast shadows in an image

A method of removing cast shadows from a digital image, the method comprising:

• identifying, by at least one processor, one or more shadow parameters for pixels in a digital image corresponding to a cast shadow;
• generating, by the at least one processor, a naïve inversion of the digital image by reducing, for each pixel corresponding to the cast shadow, one or more color space parameters of the pixel based on the one or more shadow parameters to set the one or more color space parameters at a corresponding parameter threshold;
• and
• removing the cast shadow from the digital image by synthesizing, by the at least one processor, a plurality of image patches corresponding to the cast shadow based on a plurality of non-shadow image patches from the digital image and a plurality of image patches from the naïve inversion of the digital image.

A system for identifying and removing cast shadows from a digital image, comprising:

• at least one processor;
• at least one non-transitory computer readable storage medium storing instructions thereon, that, when executed by the at least one processor, cause the system to:
• determine, for each image patch from a plurality of image patches in a digital image and using a grid-based patch-matching algorithm, a plurality of nearest neighbor patches comprising one or more visual features having a value within a predetermined range of the corresponding image patch;
• detect shadows in the digital image based on extracted feature vectors of each image patch and the corresponding nearest neighbor patches;
• identify one or more cast shadows from the detected shadows using a decision tree trained using manually labeled cast shadow regions in a training dataset;
• compare one or more color space parameters of each pixel in the one or more identified cast shadows to a corresponding parameter threshold;
• generate a naïve inversion of the digital image by modifying each pixel in the one or more identified cast shadows to meet the corresponding parameter threshold;
• and
• remove the one or more identified cast shadows by synthesizing a plurality of image patches corresponding to the one or more predicted cast shadows based on a plurality of non-shadow image patches from the digital image and the naïve inversion of the digital image.