US 8,559,672 B2Granted

Determining detection certainty in a cascade classifier

AssigneeHewlett-Packard Development Company, L.P.

InventorDarryl Greig

Life of this patent

Estimating the term…

Loading patent…

Abstract

Disclosed are embodiments for determining detection certainty in a cascade classifier (100). A first sequence (101) of detection stages (102) determines whether an image (112) is associated with a rare object. Each of the stages (102) rejects the image (112) when the image (112) is unassociated with the rare object and propagates the image (112) to a next stage (102) when the image (112) is unrejected. A second sequence (103) of detection stages (104) is configured for post-processing operation. If the image (112) is unrejected in a final stage (102d) of the first sequence (101), the second sequence (103) determines whether the image (112) is associated with the rare object. Each of the stages (104) propagates the image to a next stage (104), recording a rejection when the image (112) is unassociated with the rare object. A probability that the image (112) is associated with the rare object is determined.

Claims15

Dependent claims

A system for determining detection certainty in a cascade classifier, the system comprising:

at least one computer system;
and
a cascade classifier application executable in the at least one computer system, the cascade classifier application comprising:
a first sequence of one or more detection stages that determines whether an image is associated with a rare object, each detection stage under normal operation rejects the image when the image is determined to be unassociated with the rare object and propagates the image to a next detection stage when the image is unrejected;
a second sequence of two or more detection stages that determines whether the image is associated with the rare object responsive to the image being unrejected in a final detection stage of the first sequence, each detection stage under post-processing operation propagates the image to a next detection stage and records a rejection when the image is determined to be unassociated with the rare object, wherein each detection stage is configurable under normal or post-processing operation;
and
logic that determines a probability value that the image is associated with the rare object based at least in part on stage responses and stage decisions associated with the second sequence.

4 more elements

A method for determining detection certainty in a cascade classifier in at least one computer system, the method comprising the steps of:

determining whether an image is associated with a rare object in a first sequence of one or more detection stages, each detection stage in the first sequence rejecting the image when the image is determined by the detection stage to be unassociated with the rare object, and propagating the image to a next detection stage in the first sequence when the image is unrejected;
configuring a second sequence of two or more detection stages for post-processing operation, the detection stages of the second sequence configurable for normal or post-processing operation;
responsive to the image being unrejected in a final detection stage of the first sequence, determining whether the image is associated with the rare object in the second sequence, each detection stage in the second sequence propagating the image to a next detection stage in the second sequence, and recording a rejection when the image is determined by the detection stage to be unassociated with the rare object;
and
determining a probability value that the image is associated with the rare object based at least in part on stage responses and stage decisions associated with the second sequence.

2 more elements

Dependencies are inferred from the claim text.

Facts

Application: 13/260,354Filed 1 Jun 2009
Published: US 2012/0243734 A127 Sep 2012
Granted: 15 Oct 2013
Term ends: Not estimatedThe source records don’t support an estimate.

Classification9 codes

G06T 3/00
Geometric image transformations in the plane of the image
G06F 18/2148
Generating training patterns; Bootstrap methods, e.g. bagging or boosting: characterised by the process organisation or structure, e.g. boosting…
G06F 18/24323
Relating to the number of classes: tree-organised classifiers
G06V 10/764
Using pattern recognition or machine learning: using classification, e.g. of video objects