Digital image cropping using a blended map
View Patent ↗An improved method for cropping a main subject from a digital image is disclosed. A skin color map and face map are independently computed and then blended to produce a revised map. A region map is computed by segmenting the digital image into a plurality of regions. A main subject importance map is calculated from the revised map and the region map. The digital image is cropped based on the main subject map.
1. A method for mapping a digital image having a plurality of pixels, the method comprising the steps of:
computing a skin color map of the digital image, said skin color map having a plurality of skin color values, each of said skin color values being associated with a respective set of one or more of said pixels, each of said skin color values indicating a likelihood that the respective said pixel set is skin;
generating a face map of the digital image, said generating being independent of said computing, said face map being divided into one or more face zones and a background exclusive of said face zones, each of said face zones being a group of said pixels, each of said face zones indicating a greater likelihood than said background that the respective said group of pixels is a face;
blending the skin color map and the face map by increasing skin color values associated with the pixels of face zones providing a revised map that differentiates face skin color from other areas in the revised map;
segmenting the digital image into a plurality of regions to provide a region map, said segmenting being independent of said computing and generating;
calculating a main subject importance map from said region map and said revised map, said main subject importance map having a plurality of importance values, each of said importance values being associated with a respective region in the region map, each of said importance values indicating a likelihood that the respective said region is of importance to a main subject of the digital image;
cropping said digital image without user intervention based on said main subject importance map to provide a cropped digital image; and
outputting said cropped digital image.
2. The method of claim 1 wherein said calculating further comprises:
designating each of one or more of said regions as a skin region based on said revised map;
computing at least one geometric feature of each of said skin regions; and
wherein said importance values are based upon said revised map and respective said geometric features.
3. The method of claim 2 wherein said geometric features are selected from the group consisting of: a centrality feature and a borderness feature.
4. The method of claim 3 further comprising, prior to said cropping:
determining a main subject using said main subject importance map;
determining one or more face windows, each of said face windows being at least as large as and inclusive of a respective one or more of said face zones;
cropping said digital image to a cropping boundary inclusive of the larger of said main subject and a rectangular area inclusive of all of said face windows.