Quotient Appearance Manifold Mapping For Image Classification
AssigneeSiemens Medical Solutions USA, Inc.
InventorsYoshihisa Shinagawa, Yuping Lin, Gerardo Hermosillo Valadez, Bing Jian
Life of this patent
Estimating the term…
Abstract
Described herein is a technology for facilitating classification of an object. In one implementation, at least one quotient appearance manifold mapping is constructed from a sample image to untangle appearance fiber bundles. A feature characteristic of the sample image may then be extracted based on the quotient appearance manifold mapping. A classifier may further be trained based on the extracted feature, and adapted for associating the object with an object class.
Claims21
A computer-implemented method for classifying an object, comprising:
- (a) constructing, by a processor, at least one quotient appearance manifold mapping based on a sample image to untangle appearance fiber bundles;
- (b) reducing, by the processor, dimensionality of the quotient appearance manifold mapping to generate at least one feature characteristic of the sample image;
- and
- (c) training, by the processor, based on said feature, at least one classifier adapted for associating the object with an object class.
A computer usable medium having a computer readable program code tangibly embodied therein, said computer readable program code adapted to be executed to implement a method for classifying an object, comprising:
- (a) constructing at least one quotient appearance manifold mapping based on a sample image to untangle appearance fiber bundles;
- (b) reducing dimensionality of the quotient appearance manifold mapping to generate at least one feature characteristic of the sample image;
- and
- (c) training, based on said feature, at least one classifier adapted for associating the object with an object class.
An image analysis system, comprising:
- a memory device for storing computer readable program code;
- and
- a processor in communication with the memory device, the processor operative with the computer readable program code to
- (a) construct at least one quotient appearance manifold mapping based on a sample image to untangle appearance fiber bundles;
- (b) reduce dimensionality of the quotient appearance manifold mapping to generate at least one feature characteristic of the sample image;
- and
- (c) train, based on said feature, at least one classifier adapted for associating the object with an object class.
A method for use in a computer-aided detection system, comprising:
- (a) receiving a plurality of patches extracted from a positive class of sample images;
- (b) labeling the patches with pre-determined labels;
- (c) extracting a histograms of oriented gradients (HoG) feature of the patch;
- and
- (d) training a classifier based on the HoG feature, wherein the trained classifier is adapted to indicate a likelihood of malignancy of a lesion.
Dependencies are inferred from the claim text.