METHOD AND APPARATUS FOR INFERRING FACIAL COMPOSITE

• US 20150278997A1
• Filed: 06/24/2013
• Published: 10/01/2015
• Est. Priority Date: 09/26/2012
• Status: Active Grant

• Current Assignee: Korea Advanced Institute of Science and Technology
• Original Assignee: Korea Advanced Institute of Science and Technology
• Inventors: Kim, Ig Jae, Choi, Young Ju
• Granted Patent: US 9,691,132 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/583 using metadata automatically derived from the content
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06T 2207/10004 Still image; Photographic image
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/20221 Image fusion; Image merging
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 3/4053 based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
• CPC Class Codes: G06T 5/50 using two or more images, e.g. averaging or subtraction
• CPC Class Codes: G06V 40/171 Local features and components; Facial parts eye characteristics G06V40/18; Occluding parts, e.g. glasses; Geometrical relationships

Abstract

Provided is a method and apparatus for inferring a facial composite, whereby user'"'"'s designation regarding at least one point of a facial image is received, facial feature information are extracted based on the received user'"'"'s designation, a facial type that coincides with the extracted facial feature information is searched from a facial composite database to generate a face shape model based on the searched facial type and an initial facial composite model having a facial type similar to a face of the facial image from a low-resolution facial image through which face recognition or identification cannot be accurately performed is provided, so that the face shape model contributes to criminal arrest and a low-resolution facial image captured by a surveillance camera can be more efficiently used.

Claims

1. A method of inferring a facial composite comprising:

receiving user'"'"'s designation regarding at least one point of a facial image;

extracting facial feature information regarding a predetermined item so as to identify the facial image based on the received users designation;

searching a facial type that coincides with the extracted facial feature information using a support vector machine (SVM) from a facial composite database in which feature information according to facial type have been already learned through the SVM and are classified in a hierarchical structure having a plurality of levels and are stored; and

generating a facial composite for generating a face shape model based on the searched facial type.

2. The method of claim 1, wherein the facial feature information comprise at least one feature point according to the user'"'"'s designation and length information of segments that connect the at least one feature point.

3. The method of claim 1, wherein the generating of the facial composite comprises:

generating an initial face shape model corresponding to the searched facial type;

displaying the generated initial face shape model and receiving a users correction instruction on the displayed initial face shape model; and

completing a face shape model by correcting the initial face shape model based on the received user'"'"'s correction instruction.

4. The method of claim 1, before the receiving of the user'"'"'s designation, further comprising performing a preprocessing of removing a background light source from the facial image and increasing resolution of the facial image.

5. The method of claim 4, before the receiving of the user'"'"'s designation, further comprising determining whether the facial image having an increased resolution through the prerestoration process is successfully restored,

wherein, if it is determined that restoration of the facial image having the increased resolution fails, performing the receiving of the user'"'"'s designation.

6. An apparatus for inferring a facial composite comprising:

a user input unit receiving user'"'"'s designation regarding at least one point of a facial image;

a facial feature information extraction unit extracting facial feature information regarding a predetermined item so as to identify the facial image based on the received user'"'"'s designation;

a facial composite database in which feature information according to facial type that have been learned through a support vector machine (SVM), are classified in a hierarchical structure having a plurality of levels and are stored;

a facial composite database search unit searching a facial type that coincides with the extracted facial feature information using the SVM from the facial composite database; and

a facial composite generation unit generating a facial composite for generating a face shape model based on the searched facial type.

7. The apparatus of claim 6, wherein the facial feature information comprise at least one feature point according to the user'"'"'s designation and length information of segments that connect the at least one feature dot.

8. The apparatus of claim 6, further comprising a prerestoration unit removing a background light source from the facial image and increasing a resolution of the facial image.

9. A method of inferring a facial composite comprising:

performing a prerestoration process of removing a background light source from a facial image and of increasing a resolution of the facial image;

determining whether the facial image having an increased resolution through the prerestoration process is successfully restored;

if it is determined that restoration of the facial image having the increased resolution fails, receiving user'"'"'s designation regarding at least one point of the facial image;

extracting facial feature information regarding a predetermined item so as to identify the facial image based on the received user'"'"'s designation;

searching a facial type that coincides with the predetermined feature information according to facial type from a facial composite database in which feature information according to facial type have been already learned; and

generating a face shape model based on the searched facial type.

10. The method of claim 9, wherein the facial feature information comprise at least one feature dot according to the user'"'"'s designation and length information of segments that connect the at least one feature dot.

11. The method of claim 9, wherein the generating of the facial composite comprises:

loading an initial face shape model corresponding to the searched facial type from the facial composite database;

displaying the loaded initial face shape model and receiving user'"'"'s correction instruction on the displayed initial face shape model; and

completing a face shape model by correcting the initial face shape model based on the received user'"'"'s correction instruction.

12. The method of claim 9, wherein feature information according to facial type have been already learned through a support vector machine (SVM), are classified in a hierarchical structure having a plurality of levels and are stored in the facial composite database.

13. The method of claim 9, wherein the searching of the facial type comprises:

searching a class that belongs to the hierarchical structure of the facial composite database that coincides with the extracted facial feature information using the SVM; and

deriving a facial type corresponding to the searched class as a facial type that coincides with the extracted facial feature information.