Adaptive Scanning for Performance Enhancement in Image Detection Systems

• US 20080219558A1
• Filed: 03/09/2007
• Published: 09/11/2008
• Est. Priority Date: 03/09/2007
• Status: Active Grant

• Current Assignee: 138 East LCD Advancements Limited (Vector Capital Corporation)
• Original Assignee: Seiko Epson Corporation (Seiko Group)
• Inventors: Lu, Juwei, Zhou, Hui, Thiyagarajah, Mohanaraj
• Granted Patent: US 7,840,037 B2
• Time in Patent Office: Days
• US Class Current: 382/190
• CPC Class Codes: G06V 40/162 using pixel segmentation or colour matching
• CPC Class Codes: G06V 40/165 using facial parts and geometric relationships

Abstract

A method and system for efficiently detecting faces within a digital image. One example method includes identifying a digital image comprised of a plurality of sub-windows and performing a first scan of the digital image using a coarse detection level to eliminate the sub-windows that have a low likelihood of representing a face. The subset of the sub-windows that were not eliminated during the first scan are then scanned a second time using a fine detection level having a higher accuracy level than the coarse detection level used during the first scan to identify sub-windows having a high likelihood of representing a face.

Claims

1. A method for adaptively scanning a digital image for detecting an object, the method comprising:

identifying a digital image comprised of a plurality of sub-windows;

performing a first scan of the digital image using a coarse detection level to eliminate the sub-windows having a low likelihood of representing the object, leaving a subset of the sub-windows that have not been eliminated; and

performing a second scan of the subset of the sub-windows using a fine detection level having a higher accuracy level than the coarse detection level, to identify sub-windows having a high likelihood of representing the object.

2. The method as recited in claim 1, wherein performing the first scan of the digital image data using the coarse detection level comprises:

generating a color component for each of the sub-windows;

generating an edge component for each of the sub-windows; and

generating a simplified boosting-based component for each of the sub-windows;

wherein the sub-windows are eliminated when any of the color component, the edge component, or the simplified boosting-based component fall below predetermined levels.

3. The method as recited in claim 2, wherein generating the color component for each of the sub-windows further comprises:

identifying a sub-window of the digital image;

dividing the identified sub-window into at least two frames;

applying the at least two frames to a color map; and

generating one or more color components by summing one or more combinations of the at least two frames;

wherein the identified sub-window is eliminated when any of the one or more color components falls below a predetermined level.

4. The method as recited in claim 2, wherein generating an edge component for each of the sub-windows further comprises calculating an edge magnitude for each pixel within the digital image.

5. The method as recited in claim 2, wherein generating a simplified boosting-based component for each of the sub-windows further comprises:

identifying a sub-window of the digital image; and

passing the identified sub-window through a first portion of a plurality of classifiers, wherein each subsequent classifier has a lower false-detect rate than a previous classifier;

wherein the simplified boosting-based component falls below the predetermined level if the identified sub-window fails to pass one of the classifiers.

6. The method as recited in claim 5, wherein performing the second scan of the digital image using the fine detection level comprises:

passing the identified sub-window through a second portion of the plurality of classifiers, wherein each of the second portion of the plurality of classifiers has a lower false-detect rate than any of the first portion of the plurality of classifiers.

7. The method as recited in claim 1, further comprising adaptively updating a sub-window scale size used during the first scan and the second scan so that a greater number of sub-window scale sizes are generated near the identified sub-windows having a high likelihood of representing the object than near the sub-windows having a low likelihood of representing the object.

8. The method as recited in claim 1, further comprising processing the results of the second scan to generate an output for identifying the existence of objects within the sub-windows.

20. The method as recited in claim 1, further comprising instructions, that when executed, adaptively update a sub-window scale size used during the first scan and the second scan so that a greater number of sub-window scale sizes are generated near the identified sub-windows having a high likelihood of representing the object than near the non-face sub-windows.

9. A face detection system, comprising:

a skin color based classifier configured to analyze a digital image having a plurality of sub-windows during a first scan to eliminate the sub-windows that have a low likelihood of representing a face based on the colors of the sub-windows;

an edge based classifier coupled to the skin color based classifier and configured to analyze the digital image during the first scan to eliminate additional sub-windows that have a low likelihood of representing a face based on an edge magnitude of the digital image within the sub-windows; and

a boost classifier coupled to the edge based classifier having a plurality of cascaded modules, wherein at least a first portion of the cascaded modules are configured to analyze the digital image during the first scan to eliminate additional sub-windows that have a low likelihood of representing a face based on a first false-detect tolerance level and leaving a subset of sub-windows that have not been eliminated, and a second portion of the cascaded modules are configured to analyze during a second scan the subset of sub-windows and to eliminate additional sub-windows that have a low likelihood of representing a face based on a second false-detect tolerance level that is lower than the first false-detect tolerance level used during the first scan.

10. The face detection system as recited in claim 9, further comprising a post processor coupled to the boost classifier to process the results of the boost classifier and to generate an output for identifying the representation of faces that exist within the digital image.

11. The face detection system as recited in claim 9, wherein the skin color based classifier is further configured to generate a skin color component for each of the sub-windows, the edge based classifier is further configured to generate an edge component for each of the sub-windows, and the boost classifier is further configured to generate a simplified boosting-based component for each of the sub-windows, and wherein the sub-windows are eliminated when any of the skin color component, the edge component, or the simplified boosting-based component fall below predetermined levels.

12. The face detection system as recited in claim 9, wherein the skin color based classifier is further configured to divide each sub-window into at least two frames, apply the at least two frames to a skin-color map, and calculate the sum of one or more combinations of the at least two frames in order to eliminate the sub-window if the calculated sum falls below a predetermined level.

13. The face detection system as recited in claim 9, wherein the plurality of cascaded modules of the boost classifier are configured to analyze each of the sub-windows in successive order, and wherein each subsequent cascaded module is configured to have a lower false-detect rate than a previous cascaded module, and wherein the boost classifier is configured to reject one of the sub-windows if it fails to pass one of the cascaded modules.

14. The face detection system as recited in claim 9, wherein the first scan performed by the boost classifier is configured to have a lower computational requirement than the second scan performed by the boost classifier.

15. The face detection system as recited in claim 9, further comprising a scale size selection module configured to adaptively update a sub-window scale size used during the first scan and the second scan so that a greater number of sub-window scale sizes are generated near sub-windows that are not eliminated and a lesser number of sub-window scale sizes are generated near the sub-windows that are eliminated.

16. In a computer system that includes classifying modules for classifying an image, a computer program product configured to implement a method of adaptively scanning a digital image for face detection, the computer program product comprising one or more computer readable media having stored thereon computer executable instructions that, when executed by a processor, cause the computer system to perform the following:

identify a digital image comprised of a plurality of sub-windows;

perform a first scan of the digital image using a coarse detection level, the first scan causing the computer system to;

generate a skin color component for each of the sub-windows;

generate an edge component for each of the sub-windows; and

generate a simplified boosting-based component for each of the sub-windows;

wherein non-face sub-windows are eliminated when any of the skin color component, the edge component, or the simplified boosting-based component fall below predetermined levels, leaving a subset of the sub-windows that have not been eliminated; and

perform a second scan of the subset of the sub-windows using a fine detection level by generating a complete boosting-based component having a lower false-detect rate than the simplified boosting-based component for more accurately eliminating the non-face sub-windows and identifying sub-windows having a high likelihood of representing a face.

17. The method as recited in claim 16, wherein the computer executable instructions that generate the skin color component for each of the sub-windows further comprise instructions, that when executed:

identify a sub-window of the digital image;

divide the identified sub-window into at least two frames;

apply the at least two frames to a skin-color map; and

generate one or more skin color components by summing one or more combinations of the at least two frames;

wherein the identified sub-window is eliminated when any of the one or more skin color components falls below a predetermined level.

18. The method as recited in claim 16, wherein the computer executable instructions that generate a simplified boosting-based component for each of the sub-windows further comprise instructions, that when executed:

identify a sub-window of the digital image; and

pass the identified sub-window through a first portion of a plurality of classifiers, wherein each subsequent classifier has a lower false-detect rate than a previous classifier;

wherein the simplified boosting-based component falls below the predetermined level if the identified sub-window fails to pass one of the classifiers.

19. The method as recited in claim 18, wherein the computer executable instructions that perform the second scan of the digital image further comprise instructions, that when executed:

pass the identified sub-window through a second portion of the plurality of classifiers, wherein each of the second portion of the plurality of classifiers has a lower false-detect rate than any of the first portion of the plurality of classifiers.