Method for outlining and aligning a face in face processing of an image
First Claim
Patent Images
1. A method of training texture classifiers to detect an object in an image, the method being preformed by a computing system and comprising:
- specifying a plurality of resolution levels for an object to be detected;
specifying one or more feature points for each resolution level;
collecting positive image samples from a window centered on each feature point at each resolution level from a plurality of object images;
collecting negative image samples from windows not centered on feature points at each resolution level from a plurality of object images; and
training a texture classifier for each feature point of each resolution level using a boosting learning algorithm.
1 Assignment
0 Petitions
Accused Products
Abstract
Method and apparatus for face alignment by building a hierarchical classifier network. The hierarchical classifier network connects the tasks of face detection and face alignment into a smooth coarse-to-fine procedure. Texture classifiers are trained to recognize feature texture at different scales for different resolution layers. A multi-layer structure is employed to organize the texture classifiers, which begins with one classifier at the first layer and gradually refines the localization of feature points using additional texture classifiers in subsequent layers.
-
Citations
34 Claims
-
1. A method of training texture classifiers to detect an object in an image, the method being preformed by a computing system and comprising:
-
specifying a plurality of resolution levels for an object to be detected; specifying one or more feature points for each resolution level; collecting positive image samples from a window centered on each feature point at each resolution level from a plurality of object images; collecting negative image samples from windows not centered on feature points at each resolution level from a plurality of object images; and training a texture classifier for each feature point of each resolution level using a boosting learning algorithm. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
-
-
16. A method for detecting and aligning an object in an image, the method being performed by a computing system and comprising:
-
locating a first feature point corresponding to the object in an image using a texture classifier at a first resolution level; estimating shape and pose parameters for the object at the first resolution level; locating a set of second feature points corresponding to features of the object using texture classifiers at a second resolution level, that is finer than the first resolution level; wherein search areas for the texture classifiers at the second resolution level are constrained according to the shape and pose parameters for the object at the first resolution level. - View Dependent Claims (17, 18, 19, 20, 21, 22, 23, 24)
-
-
25. A digital computing apparatus for training texture classifiers, comprising:
-
an image input unit for inputting a plurality of images containing an object to be detected; a memory storage unit for storing positive image samples and negative image samples collected from the plurality of images containing the object; and a central processing unit for training a set of texture classifiers corresponding to a set of feature points for a plurality of resolution levels of the object using a boosting learning algorithm; wherein the positive image samples are collected by the central processing unit by specifying a plurality of resolution levels for a face, specifying a set of one or more feature points for each resolution level, and collecting positive image samples from a window centered on each feature point at each resolution level from a plurality of face images, wherein said negative image samples are collected by the central processing unit from a window positioned away from the center on each feature point at each resolution level from a plurality of face images. - View Dependent Claims (26, 27, 28)
-
-
29. A digital computing apparatus for detecting and aligning an object in an image, comprising:
-
an image input unit for inputting an image containing an object to be detected and aligned; a memory storage unit for storing texture classifiers; and a central processing unit for locating a first feature point corresponding to the object using a texture classifier at a first resolution level, estimating shape and pose parameters for the object at the first resolution level, locating a second set of feature points corresponding to features of the object using texture classifiers at a second resolution level, that is finer than the first resolution level; wherein search areas for the texture classifiers at the second resolution level are constrained according to the shape and pose parameters for the object at the first resolution level.
-
-
30. A program encoded on a computer readable medium for training texture classifiers, said program when executed by a processor causing the processor to execute the steps of:
-
specifying a plurality of resolution levels for an object to be detected; specifying one or more feature points for each resolution level; collecting positive image samples from a window centered on each feature point at each resolution level from a plurality of object images; collecting negative image samples from windows not centered on feature points at each resolution level from a plurality of object images; and training a texture classifier for each feature point of each resolution level using a boosting learning algorithm. - View Dependent Claims (31, 32, 33)
-
-
34. A program encoded on a computer readable medium for detecting and aligning an object in an image, said program when executed by a processor causing the processor to execute the steps of:
-
locating a first feature point corresponding to the object in an image using a texture classifier at a first resolution level; estimating shape and pose parameters for the object at the first resolution level; locating a second set of feature points corresponding to features of the object using texture classifiers at a second resolution level, that is finer than the first resolution level; wherein search areas for the texture classifiers at the second resolution level are constrained according to the shape and pose parameters for the object at the first resolution level.
-
Specification