System and method for controlling image quality

• US 8,620,036 B2
• Filed: 05/19/2008
• Issued: 12/31/2013
• Est. Priority Date: 05/29/2007
• Status: Active Grant

• Current Assignee: S1 Corporation (ACI Worldwide Incorporated)
• Original Assignee: S1 Corporation (ACI Worldwide Incorporated)
• Inventors: Kuznetsov, Victor D., Mun, Wang-Jin, Yang, Hae-Kwang, Lee, Yong-Jin, Irmatov, Anwar A., Buryak, Dmitry Y.
• Primary Examiner(s): Li, Ruiping
• Application Number:
  US12/123,123
• Publication Number:
  US 20080298644A1
• Time in Patent Office: 2,052 Days
• Field of Search: 382/118, 382/190, 382/206, 382/224
• US Class Current: 382/118
• CPC Class Codes: G06V 40/16 Human faces, e.g. facial parts, sketches or expressions

Abstract

An image quality control system and method is disclosed. At least one infrared camera takes a screen image of a room. When there are a plurality of cameras, images of the cameras are synchronized with respect to time, and a specific object of the image is tracked to estimate image quality of the object. When there are a plurality of cameras, a 3D screen model is reconfigured, and positions of the cameras and the infrared lighting tools are controlled. Infrared lighting and the cameras are controlled, and particularly, optical axis direction, optical magnification, exposure time, and the iris of the camera can be amended. Next, a high-quality object image list can be generated so as to process the images.

Claims

1. A method for recognizing a face within an image, comprising:

performing facial recognition of an input image using a person recognition device, with operations including;

classifying a plurality of sample images into classes, images of the plurality of sample images respectively having information on a face of a target to be recognized according to classifications;

recording classifications of the plurality of sample images in a database;

estimating similarity between the input image and the sample images in the database based on an image classification algorithm for measuring similarity of two images and determining identification thereof;

determining a first class of the input image based on a result of the estimated similarity between the input image and the sample image;

performing verification for the facial recognition of the input image using a recognition result checking device, with operations including;

generating, for each particular class of the database, a binary classifier for identifying a face of a target of the particular class of the database, wherein the binary classifier includes a multiple vector having a similarity measurement value between the input image and a corresponding sample image in the particular class of the database, and wherein the binary classifier is used to identify classes of sample images in the database that are incorrectly identified by the person recognition device;

attempting to apply the binary classifier generated for each particular class of the database to the input image;

determining a second class of the input image based on a result of successfully applying a particular binary classifier; and

combining the first and the second classes determined for the input image; and

confirming the facial recognition of the input image based on a result of the combining the first and the second classes.

2. The method of claim 1, wherein generating a binary classifier includes generating the binary classifier for suspect classifications recorded in the database, by:

generating image learning samples corresponding to face images of other persons which are different from face image samples classified in the database;

selecting a face image sample of the database based on the image learning samples;

applying a basic classification algorithm to the selected face image sample and selecting an inaccurately classified sample of the image learning samples; and

generating the binary classifier for each particular class corresponding to the inaccurately classified sample.

3. The method of claim 1, wherein the similarity measurement value between the input image and an image of the class of the input image is calculated according to a minimum distance between the input image and all the images of the class of the input image.

4. The method of claim 1, wherein attempting to apply the binary classifier generated for each particular class includes:

generating, for all the classes in the database, a distance vector between the input image and each image of a class;

applying a binary classifier of a corresponding class to the distance vector by classes; and

when one of the binary classifiers identifies the distance vector, determining that the input image belongs to a class corresponding to the binary classifier.

8. The method of claim 1, wherein applying each binary classifier applies a binary classifier to an input image whose class is not determined.

9. The method of claim 1, wherein finally determining the class of the input image includes finally determining the class of the input image to correspond to the class determined based on a result of the estimating when the class of the input image is successfully or unsuccessfully determined during the estimating.

10. The method of claim 4, wherein applying each binary classifier further includes:

when any one of the binary classifiers does not identify the distance vector, determining that the input image does not belong to any class.

5. A device for recognizing a face in an image, comprising:

an image database storing a plurality of sample images having information on a face of a target to be recognized when each of the plurality of sample images are organized by class;

a similarity measurement device configured to measure similarity of an input image and the sample images of the image database based on an image classification algorithm measuring similarity of two images and determine identification thereof;

a person recognition device configured to determine a first class of the input image based on a result of measuring similarity, the person recognition device configured to perform operations that;

classify the plurality of sample images into classes, images of the plurality of sample images respectively having information on a face of a target to be recognized according to classifications;

record classifications of the plurality of sample images in the image database;

determine a first class of the input image based on a result of the estimated similarity between the input image and the sample image, using the similarity measurement device; and

a recognition result checking device configured to attempt to apply a binary classifier for each class of the image database to identify a face of a target of a class and verify the first class of the input image, with operations that;

generate, for each particular class of the image database, the binary classifier,

wherein the binary classifier includes a multiple vector having a similarity measurement value between the input image and a corresponding sample image in the particular class of the database,

attempt to apply each binary classifier to the input image;

determine a second class of the input image based on a result of successfully applying a particular binary classifier;

combine the first and the second classes of the determined for the input image; and

finally determine which class corresponds to the input image based on a result of combining;

wherein at least one of the above devices are implemented by at least one processor.

6. The device of claim 5, wherein the recognition result checking device is configured to generate the binary classifier for suspect classifications recorded in the database, with use of:

a displayed image learning sample designer configured to generate an image learning sample corresponding to face images of persons which are different from face image samples classified in the image database;

a device configured to select a face image sample of the classes based on the image learning sample;

a device configured to apply a basic classification algorithm to the selected face image sample and selecting an inaccurately classified sample of the image learning samples; and

a binary classifier manufacturing component configured to generate the binary classifier for each particular class corresponding the inaccurately classified sample.

7. The device of claim 5, wherein the recognition result checking device:

generates, for all the classes in the image database, a distance vector between the input image and each image of a class;

applies a binary classifier of a corresponding class to the distance vector by classes; and

when one of the binary classifiers recognizes the distance vector, determines that the input image belongs to a class corresponding to the binary classifier.

11. The device of claim 5, wherein the recognition result checking device applies a binary classifier to an input image whose class is not determined by the person recognition device.

12. The device of claim 7, wherein the recognition result checking device determines the class of the input image as a class determined by the person recognition device when the class of the input image is determined by the person recognition device, and determines the class of the input image as a class determined based on the result of applying each binary classifier when the class of the input image is not determined by the person recognition device.