Method of recognizing a media item
View Patent ↗A technique for use in automated recognition of a media item involves accessing a template that includes multiple segmentation maps that each is associated with one of multiple classes to which the media item might belong. For each of the multiple classes, the segmentation map is applied to an image of the media item to extract a feature set for the image, the feature set is analyzed, and an assessment is made as to whether the media item belongs to the class.
1. A method for use in automated recognition of a media item, the method comprising:
storing in a digital memory a template that comprises multiple segmentation maps, wherein each of the segmentation maps is associated with one of multiple classes to which the media item might belong, wherein each of the segmentation maps includes segments defining one or more different features of a corresponding class, and where each of the segments is defined by clustering pixels utilizing measured intensity values of the pixels;
accessing by a computer the template; and
for each of the multiple classes:
applying by the computer the segmentation map of the class to an image of the media item to extract a feature set for the image; and
analyzing by the computer the feature set including comparing the features to the feature set and assessing whether the media item belongs to the class.
2. The method of claim 1 , further comprising concluding that the media item belongs to none of the multiple classes and, in response, concluding that the media item is not recognized.
3. The method of claim 1 , further comprising concluding that the media item belongs to one of the multiple classes and, in response, recognizing that the media item is a member of that class.
4. The method of claim 1 , further comprising, for each of the multiple classes, before applying the segmentation map:
retrieving from the template information indicating a standard image size for a media item belonging to that class; and
assessing whether the image of the media item lies within a predefined margin-of-error of the standard image size.
5. The method of claim 4 , further comprising concluding that the image of the media item lies outside the predefined margin-of-error of the standard image size for that class and, in response, concluding that the media item is not a member of that class.
6. The method of claim 4 , further comprising:
concluding that the image of the media item lies within the predefined margin-of-error of the standard image size for that class; and thereafter scaling the image of the media item to have exactly the standard image size for that class.
7. The method of claim 1 , where, for each of the multiple classes, assessing whether the media item belongs to the class includes applying the template to the feature set to calculate probability density of the media item for that class.
8. The method of claim 7 , further comprising, for each of the multiple classes, using the probability density to calculate a posterior probability that the media item belongs to the class.
9. The method of claim 8 , where, for each of the multiple classes, assessing whether the media item belongs to the class includes:
identifying one of the multiple classes for which the posterior probability is greatest;
concluding that the media item belongs to that class; and
concluding that the media item belongs to none of the other classes.
10. The method of claim 1 , where the media item is a bank note.
11. The method of claim 10 , where each of the classes corresponds to a particular currency, a particular denomination, and a particular production series to which the bank note belongs and to one of multiple orientations in which the bank note can be inserted into an automated recognition machine.
12. The method of claim 11 , wherein mean intensity values of all bank notes in a training set of N bank notes were combined to create a feature set matrix for each of the classes.
13. The method of claim 12 wherein the bank notes comprise a particular denomination and series of U.S. currency.
14. The method of claim 1 wherein clustering pixels utilizing measured intensity values of the pixels is performed by clustering all pixels having sufficiently similar measured intensity values.
15. The method of claim 14 further comprising:
determining a mean intensity value for all pixels that belong to each of said segments.
16. The method of claim 1 wherein at least one of said segments comprises multiple non-contiguous areas.