Image segmentation
Systems, methods, and computer readable media for forming a mugshot from a digital color image are provided in which a dominant face is determined using the digital color image. Person segmentation is also performed using the digital color image. An image and a mask are cropped based on the dominant face thereby forming a cropped image. Rough segmentation is performed on the cropped image. A mask is averaged in projection space based on the cropped image. The mask is refined mask and prepared for the mugshot.
1. A method of segmenting a digital image of a person, said method comprising:
detecting a face of said person, wherein said digital image is stored on a memory component;
detecting a skin of said digital image;
determining at least one body dimension of said person based on dimensions of said detected face to form at least one body map;
refining said at least one body map;
placing pixels of said refined at least one body map in a particular group of a color space by applying a Least-Square; and
outputting said at least one body map, wherein said outputting is subsequent to said refining.
2. The method as described by claim 1 , wherein said face is detected by applying a hierarchical rule.
3. The method as described by claim 1 , wherein said skin detection comprises:
generating color histograms using nonparametric approach.
4. The method as described by claim 1 , wherein said skin detection comprises:
applying a parametric approach using at least one Gaussian model.
5. The method as described by claim 1 , wherein said skin detection is pixel based.
6. The method as described by claim 1 further comprising:
generating a binary skin map by applying a threshold to probabilities associated with said skin.
7. The method as described by claim 1 , wherein said face detection comprises:
detecting one or more faces using Haar-based classifier;
resolving duplicate faces; and
in response to making no face detection, tilting said digital image.
8. The method as described by claim 1 , wherein said face detection comprises:
applying at least one of an eye map or mouth map to said digital image.
9. The method as described by claim 1 further comprising:
refining said detecting said face.
10. The method as described by claim 9 , wherein said refining comprises:
determining whether a difference between a maximum and minimum luminance within said face is greater than a certain threshold; and
in response to determining that said difference is greater than said certain threshold, confirming that said detected face is in fact a face in said digital image.
11. The method as described by claim 9 further comprising:
refining skin of said at least one body map.
12. The method as described by claim 11 , wherein said refining said skin of said at least one body map comprises:
applying gamma correction to said at least one body map.
13. The method as described by claim 1 , wherein said refining said at least one body map comprises:
identifying holes that are smaller than a given hole, wherein said given hole is determined based on a size and location of said face; and
filling said identified holes.
14. The method as described by claim 1 further comprising:
determining a dominant face in said digital image; and
cropping said digital image based on said dominant face.
15. The method as described by claim 14 , wherein said dominant face is determined based on a user input.
16. The method as described by claim 14 , wherein said determining said dominant face is based on size and centeredness of said face.
17. A non-transitory computer readable storage medium having computer-readable program code stored thereon for causing a processor to implement a method comprising:
detecting a face of a person in a digital image, wherein said digital image is stored on a memory component;
detecting a skin of said digital image;
determining at least one body dimension of said person based on dimensions of said detected face to form at least one body map;
refining said at least one body map;
placing pixels of said refined at least one body map in a particular group of a color space by applying a Least-Square; and
outputting said at least one body map, wherein said outputting is subsequent to said refining.
18. A computer system comprising:
a processor;
a display; and
a memory coupled to said processor and said display, wherein said memory comprises computer-readable program code stored thereon for causing said processor to implement a method comprising:
detecting a face of a person in a digital image stored on a memory component;
detecting a skin of said digital image;
determining at least one body dimension of said person based on dimensions of said detected face to form at least one body map;
refining said at least one body map;
placing pixels of said refined at least one body map in a particular group of a color space by applying a Least-Square; and
outputting said at least one body map, wherein said outputting is subsequent to said refining.
19. The non-transitory computer readable storage medium as described by claim 17 , wherein said face is detected by applying a hierarchical rule.
20. The computer system as described by claim 18 , wherein said skin detection is pixel based.