Face categorization and annotation of a mobile phone contact list

Abstract

A method of face categorization and annotation of a face image library includes automatically cropping a face within an acquired digital image or removing one or more non-facial items from the digital image, or both, and thereby generating a full-size face image. The full-size face image is stored with other indicia identifying a person corresponding to the face in a face image library of an embedded device such as a mobile camera phone or other handheld camera device.

Description

FIELD OF INVENTION

The invention relates to annotating images based on face regions within them.

BACKGROUND

Mobile phones and desktop applications are known to include address books. It is desired to have improved functionality in this regard for mobile devices such as mobile phones.

Usually, it is impossible or uncomfortable to bring a phone or other mobile camera device close enough to a user's face in order to fill in a frame. Thus, in many cases, a face takes a small portion of the image, and thus especially when sub-sampled as part of contact data, it can become unrecognizable. It is desired to alleviate this problem without overburdening the user.

SUMMARY OF THE INVENTION

A method is provided for face categorization and annotation of a face image library. A digital image acquisition device such as a mobile camera phone or other handheld camera device, acquires a digital image of a scene that includes a face. The face is automatically cropped or one or more non-facial items is/are removed from the digital image, or both, and a full-size face image is generated. The full-size face image is stored in a face image library along with other indicia identifying a person corresponding to the face.

The face image library may include an address book or a contact list, or both, of a mobile camera phone or other handheld camera device.

A series of preview images may be acquired, and candidate face regions may be extracted from successive frames. Location data and a cumulative confidence level that the candidate face region comprises a face may be maintained, and based on information from the series of preview images, the method may include determining that the face is present within the digital image.

Manual input of further information relating to the face may be received for storing with the full-size face image. Other indicia may be input manually by a user of the digital image acquisition device. The face may be displayed, and the user may be prompted to associate the face with the identifying indicia. A list of probable members of a contact list may be displayed, and a selection may be made from the list by the user.

The generating of the full-size face image may include building a whole face from two or more partial face images, brightening a poorly illuminated face, rotating a rotated or tilted face, correcting a red-eye, white eye or golden eye defect, and/or correcting a photographic blemish artifact within the face of the digital image, or combinations of these.

The method may include automatically transmitting the digital image to one or more persons identified within the image or to a user of the digital image acquisition device, or both.

The person identified with the face may be associated with an external device or service or both, and the digital image may be automatically transmitted to the external device or service or both.

A manual selection of a level of cropping of the face from the digital image may be made by a user.

A smile, an open eye and/or another partial face portion, may be added, from one or more stored facial images of the same identified person.

Face recognition may be applied to the face based on a library of known face images.

One or more computer readable media are also provided that are encoded with a computer program for programming one or more processors to perform any of the methods described herein.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 illustrates a method of face categorization and annotation of a face image library including cropping of the face in accordance with certain embodiments.

FIG. 2 illustrates a face annotation method involving use of reference images in accordance with certain embodiments.

FIG. 3 illustrates a face annotation method involving manual input of identifying indicia in accordance with certain embodiments.

FIG. 4 illustrates a face annotation method including image processing enhancement of the face in accordance with certain embodiments.

FIG. 5 illustrates a method of face detection and identification including automatic transmission of an acquired image in accordance with certain embodiments.

FIG. 6 illustrates a method of face categorization and annotation of a face image library including a selected level of cropping of the face in accordance with certain embodiments.

FIG. 7 illustrates a method of face annotation including replacing a portion of the face from an image store in accordance with certain embodiments.

FIG. 8 illustrates a method of face detection and face recognition in accordance with certain embodiments.

DETAILED DESCRIPTION OF THE EMBODIMENTS

An advantageous method is provided for face categorization and annotation of a face image library. A digital image acquisition device such as a mobile camera phone or other handheld camera device, acquires a digital image of a scene that includes a face. The face is automatically cropped or one or more non-facial items is/are removed from the digital image, or both, and a full-size face image is generated. The full-size face image is stored in a face image library along with other indicia identifying a person corresponding to the face.

A Contact List may be used to link known or unknown, recognized or unrecognized, faces with personal contacts. Communications may be provided for performing additional services with images. Faces may be assigned to contacts and/or a user interface, wherein image quality is improved by cropping and/or otherwise cleaning up the image, e.g., to include only faces or to make the face a certain size. A photographically captured face may be assigned to a built-in contact management system of a handheld device.

A picture may be taken, e.g., by a mobile camera-enabled device of any kind. Multiple images may also preferably be captured around the time the picture was taken which are preview, postview or reference images (together “reference images”), typically having lower resolution than the main picture. A face detection routine then finds any faces in the picture with or without the use of the reference images. The picture can be enhanced using one or more of the reference images, e.g., to add illumination, to replace a frown with a smile, to replace a blink with an open eye, or to add an otherwise occluded feature to a face. An enhanced image of a face in the picture is provided by cropping or otherwise processing the picture with or without using the reference images, and/or using the reference images to provide a better image of the face detected in the picture.

The face may be recognized or unrecognized. If it is unrecognized, then it can be added to a contacts list, along with image metadata and whatever other information a user may wish to add. If it is recognized, then the picture may be added as another look of a same person (e.g., with or without a beard, hat, glasses, certain jewelry, smiling or frowning, eyes open or blinking, one profile or the other or straight on, etc.), or just a smile, e.g., from the new picture may be added over the frown of the man picture which is otherwise kept.

A technique is provided for tracking faces in a series of images on handheld mobile devices.

In one aspect, face categorization is enhanced using a built in contact book in the phone or other mobile device.

In another aspect, a workflow and GUI are provided, wherein a user takes a picture that is associated with a name in a contact book and uses face-tracking. The phone can crop the image as well as clean it up, e.g., to keep only the face. An advantage is that with this software, saving a face to a contact becomes useful as opposed to assigning a random picture which means in some cases the face may be so small that it may not be distinguishable or resolvable. Any of the techniques described herein may be combined with those described at U.S. Pat. Nos. 6,407,777, 7,310,450, 7,315,630, 7,336,821, and 7,315,631, and US published application no. 2006/0204110, 2006/0120599, 2006/0140455, 2006/0098890, 2007/0201725, and 2007/0110305, and U.S. application Ser. Nos. 11/753,098, 11/752,925, 11/833,224, 10/763,801, 60/829,883, 11/753,397, 11/766,674, and 11/773,868, which are assigned to the same assignee and are hereby incorporated by reference.

FIG. 1 illustrates a method of face categorization and annotation of a face image library including cropping of the face in accordance with certain embodiments. A digital image is acquired at 102 including a face. The face is then automatically cropped at 104 and/or a non-facial item is removed from the digital image at 106. A full-size face image is generated at 108. The full-size face image is stored at 110 with indicia identifying a person corresponding to the face.

A digital camera may employ a face tracker which analyzes a preview image stream and extracts candidate face regions from successive frames of the preview image stream. These candidate regions are made available within the camera for additional image processing. A detailed description is given in U.S. Pat. No. 7,315,631, which is incorporated by reference. The face tracker maintains a history of each candidate face region including location data and a cumulative confidence level that the candidate region is indeed a face.

FIG. 2 illustrates a face annotation method involving use of reference images in accordance with certain embodiments. A series of reference images are acquired at 202. The reference images may be preview or post-view images, or images acquired with a different sensor than the main image at the same time or at a different time than the acquisition of the main image, at full or low resolution (see, e.g., U.S. application 60/945,558, incorporated by reference). One or more candidate face regions are extracted from successive frames at 204. Location data and a cumulative confidence level that each candidate face region comprises a face are maintained at 206. The main image is acquired at 208 including a face. Based on the reference images, it is determined at 210 that the face is present within the main image. The face is automatically cropped at 212 and/or one or more non-facial items are removed form the digital image at 214. A full-size face image is then generated at 216 based on 212 and/or 214. A full-size face image is stored with indicia indentifying a person corresponding to the face at 218.

Certain embodiments involve devices such as hand held communication devices such as mobile phones, that have a “phone book” built into the device. Face detection is tied into a process wherein a user can assign a photographically-acquired face to an existing or a new contact. Moreover, the image-processing unit can provide a saved region corresponding to the captured face, e.g., using cropping and/or removing other unnecessary details and/or by building a whole face from partial face images and/or brightening a poorly illuminated or rotated or tilted or partially-occluded face, or a face with red-eye, white eye or golden eye defects, or other blemishes possibly induced by dust artifacts in the imaging system of the camera, for what may be otherwise a good picture of a particular contact, among other processing that is possible (see cited references below).

Example

The process can be directed from the phone book or from camera software. For example:

• • 1. In the Contact manager: Create new item
  • 2. In the contact manager: option to insert new data
  • 3. In the contact manager: Option to assign picture which will then:
  • 4. let the user select: Camera or albums.
  • 5. For Camera, the user will take a picture of a subject and the camera will save the cropped image.
  • 6. For the Album, the handset will display the captured images with faces, and the user can select the right picture.
  • 7. Returning tot he contact manager: The face will be assigned to the contact.

Starting from the Camera system:

• • a. When the user grabs a picture (and optionally a face is detected) and has an option to save image or to “assign face to a contact”
  • b. The user will then select an existing contact or create a new contactto assign the face to.

The system can also be connected to a face recognition subsystem.

In other embodiments, the image acquisition appliance includes a smartphone which incorporates full mobile phone capability as well as camera capabilities. In this embodiment the recognition subsystems perform an analysis of detected face regions, and extract a pattern of DCT feature vectors, and determine if such face regions match with any of a set of “known” patterns. These “known” patterns will typically have been derived from a larger image collection stored on a user's PC or in a web service and match people in the user's friends, colleagues and family, but it may be stored on the mobile device. We remark that each person may be associated with more than one pattern and people can have different appearances at different times. If a face region matches a “known” pattern, then that face region, and the image it was extracted from, can be associated with the “known” pattern and the person that pattern is linked to. Some aspects of associating multiple face recognition patterns, or “faceprints” with individual persons, or “contacts”, are described at U.S. patent application Ser. No. 10/764,339, which is hereby incorporated by reference. Some recognition functions may be performed on a mobile device and the remainder on a desktop PC.

Initial training or learning may be performed outside the phone, e.g., in an expert system format, because better results in training and associating patterns with people can initially be achieved with larger image collections. Nevertheless it is possible to implement training, from scratch, within the device, although the process may be tedious for a casual user.

Certain embodiments provide for the creation of linkages between the known patterns of a face recognition system, and a phone contact list of a user. This can be achieved either through uploading a contact list onto a user's PC and performing association through a direct matching of face regions associated with each “known” pattern with a member of a contact list. Alternatively, it can be achieved on the phone by cycling through “known” patterns and displaying associated face regions.

In other embodiments, ease of usability of contact management is provided on a hand-held device using a built in camera.

Advantageously, a frame is cropped in certain embodiments herein to more substantially fill the frame with the face.

Furthermore, using the camera in connection with a built in contact management of a device enables use of the mobile phone as an annotation device for improving the quality of the recognition process and creating links between newly determined recognition patterns and the contact list, e.g., through a single user action. It relies on the fact that many images will either contain “unknown” people or that “known” people will occasionally be captured with a different appearance from their normal set of patterns. When such unrecognized facial regions are detected by the recognition subsystem, it displays an extracted face region on the screen and prompts the user to associate this “new” region with a member of the contact list. In alternative embodiments, the region may have a probabilistic association with members of the contact list and these may be ordered according to the determined probabilities.

FIG. 3 illustrates a face annotation method involving manual input of identifying indicia in accordance with certain embodiments. A digital image is acquired at 302 including a face. The face is then automatically cropped at 304 and/or a non-facial item is removed from the digital image at 306. A full-size face image is generated at 308. The full-size face image is displayed at 310. The user may be prompted to identify a face at 312, after which manual input may be received at 314. A list of probably members may be displayed at 316 from which a selection may be received from the list at 318. Finally, the full-size face image is stored with indicia identifying the person corresponding to the face.

The face-tracking system, discussed above with reference to FIG. 2, may also automatically rotate the image when determined to be advantageous based on the detection of a face at a particular angle. The software may allow “rotating” the face if it is not full frontal.

FIG. 4 illustrates a face annotation method including image processing enhancement of the face in accordance with certain embodiments. A digital image is acquired at 402 including a face. The face is then automatically cropped at 404 and/or a non-facial item is removed from the digital image at 406. A full-size face image is generated at 408. Several image processing options are available. A whole face may be built from two or more partial facial images at 410. A poorly illuminated face may be brightened at 412. A rotated face may be rotated right or left at 414 or up or down at 416, or in a random direction. An eye defect may be corrected at 418, such as red eye 420, white eye 422 or golden eye 424 (see U.S. Pat. Nos. 7,042,505, and 7,336,821, and US published applications 2005/0041121, 2006/0120599, and 2007/0116380, and U.S. application Ser. Nos. 11/462,035, 60/865,662, 60/892,884, 11/690,834, 11/769,206, and 11/841,855, which are hereby incorporated by reference). A photographic blemish artifact may be corrected at 426. Such blemish may be caused by an imperfection in the optical path caused by dust (see US published application 2005/0068452 and U.S. application Ser. No. 11/836,744, which are incorporated by reference). A full size face image may be stored at 428 along with identifying indicia of a person corresponding to the face.

This associating of the new face region with a member of the contact list achieves at least the following advantageous results in a single action:

It firstly associates the recognition pattern which is derived from the face region with a person in the user's contact list; this information can now be added to the set of recognition patterns and can be applied later as part of a retraining process for optimizing the set of recognition patterns associated with a user's image collection.

Another result is that is provides an association between this new image and a communications device or system with the person determined to be within the image. This could be an e-mail address or a mobile phone number. This association enables a range of added value picture services, an example of which is to enable the automated transmitting of the image to the people within the image. Faces may be found in an image, and the image may be automatically emailed to a user and/or persons associated with the faces found (see, e.g., US published patent application no. 20040243671, which is incorporated by reference), although this does not use inherent communications capabilities of the device in which the images are acquired. An enhancement of this, which relates to the “pairing mechanism” described in US published application no. 2006/0284982, which is incorporated by reference, is to provide a pairing mechanism which is triggered by selecting a member of the contact list; in this embodiment a user can associate such a member of the list with an external, networked device or service. Once such an association is established, each image which is recognized as being associated with that person can be marked for transmission to the associated device/service, placed in a transmission queue and, when the service/device next becomes available on the network, these images can be transmitted to that device/service.

FIG. 5 illustrates a method of face detection and identification including automatic transmission of an acquired image in accordance with certain embodiments. A digital image is acquired at 502 including a face. The face is then automatically cropped at 504 and/or a non-facial item is removed from the digital image at 506. A full-size face image is generated at 508. A full size face image may be stored at 510 along with identifying indicia of a person corresponding to the face. In addition, the digital image may be transmitted at 512 to an identified person or to the camera user, depending or not on whether the person is recognized in the image. A person may be associated at 514 with an external device or service. The digital image may be automatically transmitted to the external device or service at 516 depending or not on whether the person is recognized in the image.

Alternative Methods

The acquired image may be added to such database as part of the process.

In the case multiple faces are detected, a user interface may be implemented that will allow walking-through face-by-face for the user to decide if that is a face they would like to include or pass.

In a case where a camera is set in a mode of “assigning a face to a contact”, there may not be a desire to “capture” an image, but rather, the camera in preview (video) mode may continuously capture multiple images until an “acceptable image” is acquired. Such acceptable image may be a super-resolution of multiple frames, when a face is detected in frontal mode, when the image reaches focus on the face, when the light is sufficient, etc.

The process as defined herein can be extended to support Desktop based contact management software such as “ACT!” and Microsoft Outlook.

Example

• • User selects a contact;
  • User chooses the option “add image”;
  • User browses a selection of images (e.g. Folders); and
  • User selects a single image, or alternatively selects a video clip Software detects face regions; and
  • Software crops the image to include only face (and optionally face or face-and-shoulders) The software may select the level of cropping (face only, head & shoulders, etc)

FIG. 6 illustrates a method of face categorization and annotation of a face image library including a selected level of cropping of the face in accordance with certain embodiments. A digital image is acquired at 602 including a face. A manual selection of a level of cropping is received at 604 before or after the image is acquired. The face is then automatically cropped at 606. A full-size face image is generated at 608. The full size face image is stored at 610 along with identifying indicia of a person corresponding to the face.

Image processing can be added to add facial expressions such as smile. Accordingly, FIG. 7 illustrates a method of face annotation including replacing a portion of the face from an image store in accordance with certain embodiments. A digital image is acquired at 702 including a face. The face is then automatically cropped at 704. A non-facial item may be removed also at 704 or instead of cropping, and/or another face or other faces may be removed at 704. A full-size face image is generated at 706. A smile may be added from a stored image of the same person at 708, e.g., to replace a frown or occluded mouth in the image, or alternatively from a different person who may look more or less like the person. An open eye or other partial facial portion may be added at 710 from a stored image. The full size face image is stored at 710 along with identifying indicia of a person corresponding to the face.

For the recognition of known faces, the database may reside out of the handset (on the server), in case it is necessary to access a larger database than is desirable or perhaps than is possible on an handheld camera phone or other camera device.

FIG. 8 illustrates a method of face detection and face recognition in accordance with certain embodiments. A digital image is acquired at 802 including a face. The face is then automatically cropped at 804. Again, a non-facial item may be removed also at 804 or instead of cropping, and/or another face or other faces may be removed at 804. A full-size face image is generated at 806. Face recognition may be applied at 808 to the face based on a library of known face images. The full size face image is stored at 810 along with identifying indicia of a person corresponding to the face.

While an exemplary drawings and specific embodiments of the present invention have been described and illustrated, it is to be understood that that the scope of the present invention is not to be limited to the particular embodiments discussed. Thus, the embodiments shall be regarded as illustrative rather than restrictive, and it should be understood that variations may be made in those embodiments by workers skilled in the arts without departing from the scope of the present invention.

In addition, in methods that may be performed according to preferred embodiments herein and that may have been described above, the operations have been described in selected typographical sequences. However, the sequences have been selected and so ordered for typographical convenience and are not intended to imply any particular order for performing the operations, except for those where a particular order may be expressly set forth or where those of ordinary skill in the art may deem a particular order to be necessary.

In addition, all references cited herein as well as the background, invention summary, abstract and brief description of the drawings, as well as U.S. Pat. No. 6,407,777, and US published patent application nos. 20040243671 (which discloses to use faces in emails), US 20040174434 (which discloses determining meta-information by sending to a server; then back to a mobile device), 2005/0041121, 2005/0031224, 2005/0140801, 2006/0204110, 2006/0093212, 2006/0120599, and 2006/0140455, and U.S. patent application Nos. 60/773,714, 60/804,546, 60/865,375, 60/865,622, 60/829,127, 60/829,127, 60/821,165 Ser. Nos. 11/554,539, 11/464,083, 11/462,035, 11/282,954, 11/027,001, 10/764,339, 10/842,244, 11/024,046, 11/233,513, and 11/460,218, are all incorporated by reference into the detailed description of the preferred embodiments as disclosing alternative embodiments.

Claims

1. A method of face categorization and annotation of a face image library, comprising: acquiring with a digital image acquisition device a digital image of a scene that includes a face;acquiring a series of relatively low resolution reference images captured around the time of capture of said digital image;extracting candidate face regions from successive frames of said reference images;maintaining information from the successive frames of said reference images, including location data and a cumulative confidence level that the candidate face regions extracted from said successive frame of said series of reference images comprise said face, and based on the information from the series of reference images, determining that said face is present within said digital image;automatically cropping the face or removing one or more non-facial items from the digital image, or both, and thereby generating a full-size face image; andstoring the full-size face image with other indicia identifying a person corresponding to the face in a face image library.

2. The method of claim 1, wherein the face image library comprises an address book or a contact list, or both, of a mobile camera phone or other handheld camera device.

3. The method of claim 2, further comprising receiving manual input of further information relating to the face for storing with the full-size face image.

4. The method of claim 2, further comprising receiving said other indicia manually by a user of the digital image acquisition device.

5. The method of claim 4, further comprising displaying the face and prompting the user to associate the face with the identifying indicia.

6. The method of claim 5, further comprising displaying a list of probable members of a contact list and receiving a selection from the list by the user.

7. The method of claim 2, wherein the generating of the full-size face image further comprises building a whole face from two or more partial face images.

8. The method of claim 2, wherein the generating of the full-size face image further comprises brightening a poorly illuminated face, or rotating a rotated or tilted face, or combinations thereof.

9. The method of claim 2, wherein the generating of the full-size face image further comprises correcting a red-eye, white eye or golden eye defect, or combinations thereof.

10. The method of claim 2, wherein the generating of the full-size face image further comprises correcting a photographic blemish artifact within the face of the digital image.

11. The method of claim 2, further comprising automatically transmitting the digital image to one or more persons identified within the image or to a user of the digital image acquisition device, or both.

12. The method of claim 2, further comprising associating said person with an external device or service or both, and automatically transmitting said digital image to the external device or service or both.

13. The method of claim 2, further comprising receiving manual selection of a level of cropping of the face from the digital image.

14. The method of claim 2, further comprising adding a smile or open eye or other partial face portion, or combinations thereof, from one or more stored facial images of said same person.

15. The method of claim 2, further comprising applying face recognition to the face based on a library of known face images.

16. One or more non-transitory computer readable media encoded with a computer program for programming one or more processors to perform a method of face categorization and annotation of a face image library, the method comprising: acquiring with a digital image acquisition device a digital image of a scene that includes a face;acquiring a series of relatively low resolution reference images captured around the time of capture of said digital image;extracting candidate face regions from successive frames of said reference images;maintaining information from the successive frames of the reference images, including location data and a cumulative confidence level that the candidate face regions extracted from said successive frame of said series of reference images comprise said face, and based on the information from the series of reference images, determining that said face is present within said digital image;automatically cropping the face or removing one or more non-facial items from the digital image, or both, and thereby generating a full-size face image; andstoring the full-size face image with other indicia identifying a person corresponding to the face in a face image library.

17. The one or more non-transitory computer readable media of claim 16, wherein the face image library comprises an address book or a contact list, or both, of a mobile camera phone or other handheld camera device.

18. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises receiving manual input of further information relating to the face for storing with the full-size face image.

19. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises receiving said other indicia manually by a user of the digital image acquisition device.

20. The one or more non-transitory computer readable media of claim 19, wherein the method further comprises displaying the face and prompting the user to associate the face with the identifying indicia.

21. The one or more non-transitory computer readable media of claim 20, wherein the method further comprises displaying a list of probable members of a contact list and receiving a selection from the list by the user.

22. The one or more non-transitory computer readable media of claim 17, wherein the generating of the full-size face image further comprises building a whole face from two or more partial face images.

23. The one or more non-transitory computer readable media of claim 17, wherein the generating of the full-size face image further comprises brightening a poorly illuminated face, or rotating a rotated or tilted face, or combinations thereof.

24. The one or more non-transitory computer readable media of claim 17, wherein the generating of the full-size face image further comprises correcting a red-eye, white eye or golden eye defect, or combinations thereof.

25. The one or more non-transitory computer readable media of claim 17, wherein the generating of the full-size face image further comprises correcting a photographic blemish artifact within the face of the digital image.

26. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises automatically transmitting the digital image to one or more persons identified within the image or to a user of the digital image acquisition device, or both.

27. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises associating said person with an external device or service or both, and automatically transmitting said digital image to the external device or service or both.

28. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises receiving manual selection of a level of cropping of the face from the digital image.

29. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises adding a smile or open eye or other partial face portion, or combinations thereof, from one or more stored facial images of said same person.

30. The one or more non-transitory computer readable media of claim 17, wherein the method further comprises applying face recognition to the face based on a library of known face images.