Methods and apparatus for providing a scalable identification of digital video sequences

1. A method for content based video database formation, the method comprising:

segmenting selected video frames utilizing selected active frame areas into a plurality of regions, each region delineated by computed contours, wherein the active video frame areas are selected based on video filtering and temporal statistic collection of a video sequence;

tracing the computed contours to produce a vector of contiguous spatial coordinates for each region;

processing each vector of contiguous spatial coordinates to provide a contour pixel gradient vector orientation and magnitude for selected contours;

generating for each of the selected contours a signature based on the contour pixel gradient vector orientation and magnitude; and

forming a content based video database to hold the signatures, wherein the step of processing the vector of contiguous spatial coordinates comprises:

computing a normalized intensity, gradient (NIG) value for each computed contour; and

pruning each computed contour having an NIG value less than an NIG threshold to produce a set of computed contours having high NIG values.

2. The method of claim 1 , wherein the step of pruning each computed contour comprises:

adjusting the NIG threshold used in an automatic thresholding function based on a sum of lower NIG values up to a desired percentage of a total NIG sum for each video frame.

3. The method of claim 1 , wherein the step of generating for each of the selected contours a signature comprises:

selecting image contour prominent p i points and drawing vectors originating from each p i point to other prominent q, points on each of the selected contours; and

deriving a histogram for each selected prominent p i point using bins in a functional space pointed by the vectors originating from the selected prominent p i point, wherein the bins hold a selected number of contour pixels falling in the bin area.

4. The method of claim 3 further comprising:

drawing a tangent line through each of the selected image contour prominent p i points;

determining an orientation of a log-polar space accommodating an associated video frame active region; and

selecting half of this space containing the majority of the contour points for further processing to effectively reduce in half the total number of bins used for counting contour pixels falling into selected bins.

5. The method of claim 3 further comprising:

generating sector dividing lines to form vectors which begin from the selected image contour prominent p i points to selected contour prominent points to form counting bins at equidistant angles.

6. The method of claim 1 further comprising:

generating for each of the selected contours a signature based on the a histogram of contour pixel gradient vector orientations and gradient magnitudes.

7. The method of claim 1 further comprising:

selecting prominent pixels on the computed contours;

drawing a plurality of concentric circles surrounding each of the selected prominent pixels, wherein each of the largest concentric circles has a radius based on the farthest pixel on the computed contour from the associated selected prominent pixel; and

dividing the concentric circles into equal distance sectors based on vectors drawn from the selected prominent pixels to other prominent contour points to form counting bins as areas bounded by the concentric circles and the sector dividing lines.

8. The method of claim 7 further comprising:

drawing a tangent line through each of the selected prominent pixels based on an angle of a directional gradient computed for each of the selected prominent pixels; and

determining a functional space orientation and which half of the set of concentric circles to use for histogram bin formation based on a majority of contour pixels falling on one or the other side of a tangent line.

9. The method of claim 8 further comprising:

generating a contour shape descriptor by counting how many contour pixels fall into each counting bin; and

applying a normalized threshold to the contour shape descriptor to account for each bin size and to derive a binary value for each bin.

10. The method of claim 8 further comprising:

weighing all contour pixels in the counting bins based on properties of the contour;

applying a threshold normalized with respect to bin area to the number of contour pixels in each bin to generate a binary value of 0 or 1, wherein the set of 0s and 1 s corresponding to each counting bin of the log-polar grid form a multidimensional signature for the particular frame.

11. A method for content based video database formation, the method comprising:

segmenting selected video frames utilizing selected active frame areas into a plurality of regions, each region delineated by computed contours, wherein the active video frame areas are selected based on video filtering and temporal statistic collection of a video sequence;

tracing the computed contours to produce a vector of contiguous spatial coordinates for each region;

processing each vector of contiguous spatial coordinates to provide a contour pixel gradient vector orientation and magnitude for selected contours;

generating for each of the selected contours a signature based on the contour pixel gradient vector orientation and magnitude; and

forming a content based video database to hold the signatures, wherein the step of processing the vector of contiguous spatial coordinates comprises:

computing an orientation energy gradient (FOG) value for each computed contour; and

pruning each computed contour having an EOG value greater than an EOG threshold to produce a set of computed contours having low EOG values.

12. The method of claim 11 , wherein the step of pruning each computed contour comprises:

adjusting an automatic thresholding function based on a sum of higher EOG values up to a desired percentage of a total EOG sum for each video frame.

13. The method of claim 11 further comprising:

generating a spatial derivative based on an eight pixel neighborhood for contour pixels in selected contours;

generating contour pixel gradient vector orientation feature vectors based on the spatial derivative; and

generating contour pixel gradient magnitude feature vectors based on the spatial derivative.

14. The method of claim 11 further comprising:

weighing contour pixel gradient vector orientation feature vectors based on properties of selected contours;

weighing contour pixel gradient magnitude feature vectors based on the properties of the selected contours;

applying a first threshold normalized with respect to bin area to the weighted contour pixel gradient vector orientation feature vectors to generate a first set of multidimensional signatures for the particular frame; and

applying a second threshold normalized with respect to the bin area to the weighted contour pixel gradient magnitude feature vectors to generate a second set of multidimensional signatures for the particular frame, wherein the first set of multidimensional signatures and the second set of multidimensional signatures are stored in a signature library to form a video database.

15. A method for content based video database formation, the method comprising:

segmenting selected video frames into a plurality of regions, each region delineated by computed contours:

tracing the computed contours to produce a vector of contiguous spatial coordinates for each region;

processing each vector of contiguous spatial coordinates to provide a contour pixel gradient vector orientation and magnitude for contour pixels in selected contours;

generating for each of the selected contours a plurality of signatures based on the contour pixel gradient vector orientation and magnitude;

drawing a rectangular grid to form counting bins;

weighing all contour pixels in the counting bins based on properties of the selected contours;

applying a threshold normalized with respect to bin area to the number of contour pixels in each bin to generate a binary value of 0 or 1, wherein the set of 0s and 1s corresponding to each counting bin of the rectangular grid form a multidimensional signature for the particular frame; and

forming a content based video database to hold the plurality of signatures.

16. The method of claim 15 , wherein the step of processing each vector of contiguous spatial coordinates comprises:

computing a normalized intensity gradient (NIG) value for each of the computed contours; and

pruning each computed contour having an NIG value less than an NIG threshold to produce a set of computed contours having high NIG values.

17. A method for content based video database formation, the method comprising:

segmenting selected video frames into a plurality of regions, each region delineated by computed contours;

tracing the computed contours to produce a vector of contiguous spatial coordinates for each region;

processing each vector of contiguous spatial coordinates to provide a contour pixel gradient vector orientation and magnitude for contour pixels in selected contours;

generating a spatial derivative based on an eight pixel neighborhood for each contour pixel in the selected contours;

generating contour pixel gradient vector orientation feature vectors based on the spatial derivative;

generating contour pixel gradient magnitude feature vectors based on the spatial derivative;

generating for each of the selected contours a plurality of signatures based on the contour pixel gradient vector orientation and magnitude; and

forming a content based video database to hold the plurality of signatures.

18. The method of claim 17 further comprising:

weighing contour pixel gradient vector orientation feature vectors based on properties of the selected contours;

weighing contour pixel gradient magnitude feature vectors based on the properties of the selected contours;

applying a first threshold normalized with respect to bin area to the weighted contour pixel gradient vector orientation feature vectors to generate a first set of multidimensional signatures for the particular frame; and

applying a second threshold normalized with respect to the bin area to the weighted contour pixel gradient magnitude feature vectors to generate a second set of multidimensional signatures for the particular frame.

19. The method of claim 18 , wherein the first set of multidimensional signatures and the second set of multidimensional signatures are stored in a signature library to form a video database.