Method and apparatus for cancelling lighting variations in object recognition

• US 6,701,026 B1
• Filed: 04/07/2000
• Issued: 03/02/2004
• Est. Priority Date: 01/26/2000
• Status: Expired due to Fees

• Current Assignee: Kent Ridge Digital Labs
• Original Assignee: Kent Ridge Digital Labs
• Inventors: Zheng, Lei, Wu, Jian Kang
• Primary Examiner(s): Patel, Jayanti K.
• Assistant Examiner(s): CHAWAN, SHEELA C
• Application Number:
  US09/545,885
• Time in Patent Office: 1,425 Days
• Field of Search: 382/274, 382/167, 382/173, 382/254, 382/266, 382/275, 382/279, 382/260, 345/302, 345/473, 706/14, 358/450, 358/474
• US Class Current: 382/274
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20021 Dividing image into blocks, subimages or windows
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30201 Face
• CPC Class Codes: G06T 5/94 based on local image properties, e.g. for local contrast enhancement
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 10/60 relating to illumination properties, e.g. using a reflectance or lighting model

Abstract

A lighting normalisation method, apparatus and computer program product are disclosed. First, lighting parameters are estimated (110) optimally based on a shading model. After applying segmentation (120), each generated region is assigned are liability based on the stochastic model of that object. This reliability is used to distinguish (130) shadow from object detail. Regions are then further split and merged (140) according to these reliabilities. Finally, the lighting parameters of these regions are calculated and the object is reconstructed (150) to a normalised lighting condition. This technique can be used for an object recognition system, i.e., face recognition applications, as well as for image enhancement, image retrieval, machine vision and related regions.

Claims

1. A method of cancelling lighting variations in a digital image of a specified class of objects, said method including the steps of:

estimating lighting parameters and reliability for each pixel of said digital image using a stochastic model of the specified object class;

segmenting said digital image into regions having different lighting parameters;

distinguishing object details in said digital image from shadows in said digital image;

splitting and merging regions containing object details into nearest regions;

estimating predetermined lighting-parameter characteristics for each split and merged region; and

reconstructing said digital image based on said estimated lighting-parameter characteristics.

2. The method according to claim 1, further including the step of training said stochastic model of the specified object class.

3. The method according to claim 1, wherein said predetermined lighting-parameter characteristics are lighting parameters.

4. The method according to claim 1, wherein said predetermined lighting-parameter characteristics are lighting parameter functions.

5. The method according to claim 1, wherein said stochastic model is designated by or induced from a priori knowledge.

6. The method according to claim 1, wherein said steps of estimating lighting parameters use predefined blocks of predetermined size and shape.

7. The method according to claim 1, wherein said predefined blocks include lines having a length and orientation.

8. The method according to claim 1, further including the step of eliminating highlight effects from said digital image.

9. An apparatus for cancelling lighting variations in a digital image of a specified class of objects, said apparatus including:

means for estimating lighting parameters and a reliability for each pixel of said digital image using a stochastic model of the specified object class;

means for segmenting said digital image into regions having different lighting parameters;

means for distinguishing object details in said digital image from shadows in said digital image;

means for splitting and merging regions containing object details into nearest regions;

means for estimating predetermined lighting-parameter characteristics for each split and merged region; and

means for reconstructing said digital image based on said estimated lighting-parameter characteristics.

10. The apparatus according to claim 9, further including means for training said stochastic model of the specified object class.

11. The apparatus according to claim 9, wherein said predetermined lighting-parameter characteristics are lighting parameters.

12. The apparatus according to claim 9, wherein said predetermined lighting-parameter characteristics are lighting parameter functions.

13. The apparatus according to claim 9, wherein said stochastic model is designated by or induced from a priori knowledge.

14. The apparatus according to claim 9, wherein both means for estimating lighting parameters use predefined blocks of predetermined size and shape.

15. The apparatus according to claim 9, wherein said predefined blocks include lines having a length and orientation.

16. The apparatus according to claim 9, further including means for eliminating highlight effects from said digital image.

17. A computer program product having a computer usable medium having a computer readable program code means embodied therein for cancelling lighting variations in a digital image of a specified class of objects, said computer program product including:

computer readable program code means for estimating lighting parameters and a reliability for each pixel of said digital image using a stochastic model of the specified object class;

computer readable program code means for segmenting said digital image into regions having different lighting parameters;

computer readable program code means for distinguishing object details in said digital image from shadows in said digital image;

computer readable program code means for splitting and merging regions containing object details into nearest regions;

computer readable program code means for estimating predetermined lighting-parameter characteristics for each split and merged region; and

computer readable program code means for reconstructing said digital image based on said estimated lighting-parameter characteristics.

18. The computer program product according to claim 17, further including computer readable program code means for training said stochastic model of the specified object class.

19. The computer program product according to claim 17, wherein said predetermined lighting-parameter characteristics are lighting parameters.

20. The computer program product according to claim 17, wherein said predetermined lighting-parameter characteristics are lighting parameter functions.

21. The computer program product according to claim 17, wherein said stochastic model is designated by or induced from a priori knowledge.

22. The computer program product according to claim 17, wherein both computer readable program code means for estimating lighting parameters use predefined blocks of predetermined size and shape.

23. The computer program product according to claim 17, wherein said predefined blocks include lines having a length and orientation.

24. The computer program product according to claim 17, further including computer readable program code means for eliminating highlight effects from said digital image.