System and method for face detection using face region location and size predictions and computer program product thereof

• US 8,559,707 B2
• Filed: 07/12/2011
• Issued: 10/15/2013
• Est. Priority Date: 12/15/2010
• Status: Active Grant

• Current Assignee: Industrial Technology Research Institute
• Original Assignee: Industrial Technology Research Institute
• Inventors: Wu, Yi-Ta
• Primary Examiner(s): Sherali, Ishrat I
• Application Number:
  US13/180,804
• Publication Number:
  US 20120155757A1
• Time in Patent Office: 826 Days
• Field of Search: 382/103, 382/115, 382/118, 382164-165, 382170-171, 382180-181, 382/276, 382/289, 382/294, 348169-172
• US Class Current: 382/165
• CPC Class Codes: G06V 10/25 Determination of region of interest [ROI] or a volume of interest [VOI]
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

In one exemplary embodiment, a face detection method implemented in a computer system estimates location and size of face candidate regions for an input image, which includes image regionalization, region labeling and feature capturing, distance transform, and face location and size estimation, to predict at least a face location and size range for the input image. According to the at least a predicted face location and size range, the method further performs an adaptive face detection until a predicted face region is verified.

Claims

1. A method for face detection using face region location and size predictions, applicable to a computer system, said method comprising the steps of:

performing face candidate region location and size predictions on an input image, which including image regionalization, region labeling and feature capturing, distance transform with constraint, and face location and size estimation, to predict at least a face location and size range for said input image; and

according to said at least a predicted face location and size range, performing an adaptive face detection, further comprising;

establishing face detection range, establishing window size range used for face detection, and dynamically determining at least an optimal face location for each scale of a pre-defined number of scales until at least an optimal face region being verified as face location or all scales being verified, and outputting a final face detection result.

2. The method as claimed in claim 1, wherein said input image is one of grayscale image or color image.

3. The method as claimed in claim 1, wherein said image regionalization further includes:

using a clustering method to cluster a histogram corresponding to said input image and then restores said histogram clustering result to an original source image.

4. The method as claimed in claim 3, wherein when said input image is a grayscale image, said histogram is a 1-dimensional histogram.

5. The method as claimed in claim 3, wherein when said input image is a color image, said histogram is an at least 1-dimensional histogram.

6. The method as claimed in claim 1, wherein said region labeling and feature capturing further includes:

applying a connected component labeling scheme to a plurality of divided regions after the image regionalization, and assigning a specific label to each of at least a connected region of the plurality of divided regions; and

computing at least a feature of the connected region of said at least a connected region.

7. The method as claimed in claim 6, wherein said distance transform with constraint further includes:

establishing a new mask image according to said plurality of divided regions after the image regionalization, and said specific label and said at least a feature of each connected region after the region labeling and feature capturing; and

performing a distance transform on said new mask image to generate a distance map.

8. The method as claimed in claim 7, wherein said distance map represents shortest distance between each pixel of a foreground object of said input image and its background pixel of said input image.

9. A system for face detection using face region location and size predictions, said system comprising:

a processor, performing face candidate region location and size predictions on an input image, which further including image regionalization, region labeling and feature capturing, distance transform with constraint, and face location and size estimation; and

a detector, according to at least a predicted face location and size range, performing an adaptive face detection until predicted face region being verified and a final face detection result being outputted.

10. The system as claimed in claim 9, wherein said processor uses an image regionalization module to perform image regionalization on said input image by dividing said input image into a plurality of regions.

11. The system as claimed in claim 10, wherein said processor uses a region labeling and feature capturing module to apply a connected component labeling scheme to assign a specific label to each of at least a connected region of a regionalized image after the image regionalization, and compute at least a feature of said connected region of said at least a connected region.

12. The system as claimed in claim 11, wherein said processor uses a distance transform with constraint module to establish a new mask image according to said plurality of regions after the image regionalization, and said specific label and said at least a feature of each connected region of said at least a connected region after the region labeling and feature capturing, and to perform a distance transform on said new mask image to generate a distance map.

13. The system as claimed in claim 12, wherein said processor uses a face location and size estimation module to analyze said distance map to predict face location and size range of said input image.

14. A computer program product, comprising a memory and an executable computer program stored in said memory, said computer program including instructions to perform face candidate region location and size estimation and adaptive face detection, and said instructions being executed as follows:

performing, by using a processor, face candidate region location and size predictions on an input image, which including image regionalization, region labeling and feature capturing, distance transform with constraint, and face location and size estimation, to predict at least a face location and size range for said input image; and

according to said at least a predicted face location and size range, performing, by using said processor, an adaptive face detection until a predicted face region being verified and a final face detection result being outputted.

15. The computer program product as claimed in claim 14, wherein said adaptive face detection further includes:

establishing face detection range, establishing window size range used for face detection, and dynamically determining at least an optimal face location for each scale of a pre-defined number of scales.