Automatically mining person models of celebrities for visual search applications
Methods and systems for automated identification of celebrity face images are provided that generate a name list of prominent celebrities, obtain a set of images and corresponding feature vectors for each name, detect faces within the set of images, and remove non-face images. An analysis of the images is performed using an intra-model analysis, an inter-model analysis, and a spectral analysis to return highly accurate biometric models for each of the individuals present in the name list. Recognition is then performed based on precision and recall to identify the face images as belonging to a celebrity or indicate that the face is unknown.
1. A computer-implemented method of automatic face recognition, comprising:
obtaining a name list from one or more articles, wherein the name list includes a name for each person in a group of persons;
obtaining a collection of images for each person in the group using the name of that person;
selecting, from the collection of images for each person in the group, a set of representative images for that person, each of the representative images being deemed as having a highest score amongst the collection with regard to a depiction of a headshot of that person;
wherein selecting the set of representative images includes:
determining a first set of one or more images from the collection which are deemed duplicative based on comparing images of the collection with other images of the collection;
determining a second set of one or more images from the collection which match to an image in a collection of images for another person in the group; and
determining, from a remainder of the collection which excludes the first set and the second set, one or more images which are considered to be outliers as compared to a remainder of images in the collection, in order to select the set of representative images.
2. The computer-implemented method of claim 1 , further comprising filtering the one or more articles to retain only articles that contain names of people.
3. The computer-implemented method of claim 1 , wherein selecting the set of representative images includes performing iterative binary clustering.
4. The computer-implemented method of claim 1 , wherein determining the first set of one or more images from the collection includes performing intra-model analysis, and wherein determining the second set of one or more images from the collection includes performing inter-model analysis, and wherein performing the inter-model analysis is performed after the intra-model analysis.
5. The computer-implemented method of claim 1 , further comprising ranking a set of names from the name list based on a quantity in the set of representative images of each person associated with one of the names in the set of names.
6. The computer-implemented method of claim 1 , further comprising detecting a feature vector for one or more images in the set of representative images of the one or more persons in the group.
7. The computer-implemented method of claim 6 , wherein detecting the feature includes identifying a facial feature location within the one or more images in the set of representative images of the one or more persons in the group.
8. A system, comprising:
(a) a face image database;
(b) a name database; and
(c) a computer-based face recognition system, comprising:
(i) a name list generator to obtain a name list from one or more articles, wherein the name list includes a name for each person in a group of persons;
(ii) a face signature detector to obtain a collection of images for each person in the group;
(iii) one or more analyzers to select, from the collection of images for each person in the group, a set of representative images for that person, each of the representative images being deemed as having a highest score amongst the collection with regard to a depiction of a headshot of that person;
wherein the one or more analyzers select the set of representative images by:
determining a first set of one or more images from the collection which are deemed duplicative based on comparing images of the collection with other images of the collection,
determining a second set of one or more images from the collection which match to an image in a collection of images for another person in the group, and
determining, from a remainder of the collection which excludes the first set and the second set, one or more images which are considered to be outliers as compared to a remainder of images in the collection, in order to select the set of representative images.
9. The system of claim 8 , further comprising a recognizer configured to determine whether an input image depicts a person associated with a particular name in the name list based on the set of representative images for that person.
10. The system of claim 8 , wherein the name list generator further comprises a name ranker configured to rank the one or more names in the name list based on a quantity in the set of representative images of each person in the name list.
11. The system of claim 8 , wherein the face signature detector obtains the collection of images for each person in the group using a feature detector to detect face images based on Gabor wavelets.
12. The system of claim 8 , wherein the face signature detector obtains the collection of images for each person in the group using a feature detector to detect face images based on a facial feature location within the one or more face images.
13. The system of claim 9 , wherein the recognizer determines that there is no matching name associated with a given image.
14. The computer-implemented method of claim 1 , further comprising performing recognition of an input image using the set of representative images for one or more of the persons in the group.
15. A non-transitory computer-readable medium that stores instructions, including instructions that when executed by one or more processors, cause the one or more processors to perform operations that comprise:
obtaining a name list from one or more articles, wherein the name list includes a name for each person in a group of persons;
obtaining a collection of images for each person in the group using the name of that person;
selecting, from the collection of images for each person in the group, a set of representative images for that person, each of thSe representative images being deemed as having a highest score amongst the collection with regard to a depiction of a headshot of that person;
wherein selecting the set of representative images includes:
determining a first set of one or more images from the collection which are deemed duplicative based on comparing images of the collection with other images of the collection;
determining a second set of one or more images from the collection which match to an image in a collection of images for another person in the group; and
determining, from a remainder of the collection which excludes the first set and the second set, one or more images which are considered to be outliers as compared to a remainder of images in the collection, in order to select the set of representative images.
16. The non-transitory computer-readable medium of claim 15 , further comprising instructions for filtering the one or more articles to retain only articles that contain names of people.
17. The non-transitory computer-readable medium of claim 15 , wherein instructions for determining the set of representative images includes instructions for performing iterative binary clustering.
18. The non-transitory computer-readable medium of claim 15 , wherein instructions for determining the first set of one or more images from the collection includes instructions for performing intra-model analysis, and wherein instructions for determining the second set of one or more images from the collection includes instructions for performing inter-model analysis, and wherein performing the inter-model analysis is performed after the intra-model analysis.
19. The non-transitory computer-readable medium of claim 15 , further comprising instructions for ranking a set of names from the name list based on a quantity of associated representative images of each person associated with one of the names in the set of names.
20. The non-transitory computer-readable medium of claim 15 , further comprising instructions for performing recognition of an input image using the set of representative images for one or more of the persons in the group.