Face detection method and apparatus
Image fragments are formed in regions corresponding to circles searched from an input image. In a cascade of homogeneous classifiers, each classifier classifies input vectors corresponding to the image fragments into a face type and a non-face type. This procedure is performed on all images included in an image pyramid and the coordinates of a face detected based on the results of the procedures on all images.
1. A method for detecting a face, comprising:
detecting a circle having a predetermined radius from an input image by using a Hough transform;
dividing a region of the detected circle to form image fragments of the same size;
generating input vectors for each of the image fragments by using each classifier of a cascade that includes homogeneous classifiers, wherein input vectors are calculated based on local binary descriptors and the local binary descriptors include a pairwise intensity comparison between a pixel and eight or fewer neighboring pixels;
when an input vector generated for each of the image fragments is classified into a non-face type, determining that no face is detected in the input image; and
when an input vector generated for each of the image fragments is classified into a face type by the classifier of the cascade, determining that a face is detected from the input image.
2. The method of claim 1 , the detecting of a circle further includes forming an image pyramid including images that are copies of images scaled down by a predetermined ratio.
3. The method of claim 2 , wherein the classifying of input vectors further includes
training the homogeneous classifiers based on a database including a training set.
4. The method of claim 1 , wherein the classifier is provided with an input vector of a larger size in comparison with a previous classifier, and
the input vector is formed for an image fragment corresponding to an image scaled down by a predetermined ratio, and the predetermined ratio gradationally is reduced.
5. The method of claim 1 , wherein the classifier of the cascade operates based on an artificial neural network of a spare network of winnows architecture (SNOW structure).
6. An apparatus for detecting a face, comprising:
a cascade classifier that is formed in a type of a cascade, includes homogeneous classifiers, and classifies input vectors into a face type and a non-face type based on input images;
a circle detector that detects a circle having a predetermined radius from the input image by using a Hough transform; and
an image analyzer that forms image fragments of the same size within a region of the detected circle, wherein the image analyzer generates input vectors for the image fragments by using the classifiers of the cascade classifier, wherein input vectors are calculated based on local binary descriptors and the local binary descriptors include a pairwise intensity comparison between a pixel and eight or fewer neighboring pixels, determines that no face is detected from the input image when the input vector is classified into the no-face type by the classifiers, and determines that a face is detected from the input image when the input vector is classified into the face type by all the classifiers.
7. The apparatus of claim 6 , further comprising
an image pyramid constructor for including images that are copies of images scaled by a predetermined ratio.
8. The apparatus of claim 6 , further comprising
a training unit for training the homogeneous classifiers based on a database including a training set.
9. The apparatus of claim 6 , wherein the classifier is provided with an input vector of a larger size in comparison with a previous classifier, and
the input vector is formed for an image fragment corresponding to an image scaled down by a predetermined ratio, and the predetermined ratio gradationally is reduced.
10. The apparatus of claim 6 , wherein the classifier of the cascade operates based on an artificial neural network of a spare network of winnows architecture (SNOW structure).
11. The apparatus of claim 6 , further comprising a face integrator that calculates coordinates of the face detected by the image analyzer.