Back to results
US 7,548,637 B2Granted
Method for detecting objects in an image using pair-wise pixel discriminative features
AssigneeThe Board of Trustees of the University of Illinois
InventorsZiyou Xiong, Thomas S. Huang, Makoto Yoshida
Life of this patent
Estimating the term…
Abstract
A method for detecting an object in an image includes calculating a log L likelihood of pairs of pixels at select positions in the image that are derived from training images. The calculated log L likelihood of the pairs of pixels is compared with a threshold value. The object is detected when the calculated log L likelihood is greater than the threshold value.
Claims17
A method for detecting an object in an image comprising:
- calculating, using a processor, a log L likelihood of pairs of pixels at select positions in the image, said select positions being derived from a plurality of first training images and a plurality of second training images;
- comparing, using a processor, said calculated log L likelihood of said pairs of pixels with a threshold value;
- and
- determining, using a processor, that the object has been detected when said calculated log L likelihood is greater than said threshold value;
- wherein said deriving of said select positions in the image includes,
- calculating pair-wise conditional distributions corresponding to the plurality of said first training images having a predetermined dimension and sample objects and said plurality of second training images having said predetermined dimension and sample non-objects;
- calculating conditional relative entropy values between all pairs of pixels in said first and second training images based on said pair-wise conditional distributions;
- and
- selecting a plurality of the largest conditional relative entropy values and corresponding pairs of pixels from said conditional relative entropy values calculated for said all pairs of pixels;
- and
- wherein said select positions correspond to said selected pairs of pixels from said conditional relative entropy values calculated for said all pairs of pixels.
8 more elements
A method for detecting a face in an image, comprising:
- calculating, using a processor, a log L likelihood of pairs of pixels at select positions in the image, said select positions being derived from a plurality of first training images and a plurality of second training images;
- comparing, using a processor, said calculated log L likelihood of said pairs of pixels with a threshold value;
- and
- determining, using a processor, that the face has been detected when said calculated log L likelihood is greater than said threshold value;
- wherein said deriving of said select positions in the image includes,
- calculating histograms corresponding to a plurality of training face images having a predetermined dimension and a plurality of training non-face images having said predetermined dimension, said histograms being defined by probability and intensity of pixels in said training face and non-face images;
- calculating Kullback-Leibler distances (D) between all pairs of pixels in said training face images and said training non-face images based on said histograms;
- and
- selecting a plurality of largest Kullback-Leibler distance (D) values and corresponding pairs of pixels from said Kullback-Leibler distances (D) calculated for all said pairs of pixels;
- and
- wherein said select positions correspond to said selected pairs of pixels from Kullback-Leibler distance values calculated for said all pairs of pixels.
8 more elements
A method for detecting a face in an image, comprising:
- calculating, using a processor, a log L likelihood of pairs of pixels at select positions in the image, said select positions being derived from a plurality of first training images and a plurality of second training images;
- comparing, using a processor, said calculated log L likelihood of said pairs of pixels with a threshold value;
- determining, using a processor, that the face has been detected when said calculated log L likelihood is greater than said threshold value;
- and
- reducing a test image into a plurality of smaller test images, and calculating said log L likelihood with respect to pixels in each of said plurality of smaller test images;
- wherein the face is determined to be detected in the test image when said calculated log L likelihood of at least one of said smaller test images is greater than said threshold value.
3 more elements
Dependencies are inferred from the claim text.