Heuristic method for scene cut detection in digital baseband video

1. A video detector for detecting scene changes in a video that may include a number of different scenes, the video including a plurality of adjacent video frames each made from individual pixels, the detector comprising:

an input for accepting the video;

a difference metric calculator for computing a difference metric between pairs of adjacent video frames in the plurality of adjacent video frames;

a block mismatch calculator configured to compare each of a plurality of two-dimensional blocks of one of the pairs of adjacent frames to a set of corresponding blocks of a plurality of two-dimensional blocks of a second of the pairs of adjacent frames, the block mismatch calculator further configured to compute, for the pairs of adjacent frames, a percentage of block mismatch based on the comparisons and a total number of blocks in the first of the pairs of adjacent frames;

an outlier detector configured to identify a Gaussian distribution for the computed difference metrics and to identify a Gaussian distribution for the computed percentages of block mismatch, the outlier detector also configured to detect whether a difference metric outlier exists, the difference metric outlier existing when one of the computed difference metrics is outside of a threshold level of standard deviations of the Gaussian distribution for the computed difference metrics, the outlier detector further configured to detect whether a block mismatch outlier exists, the block mismatch outlier existing when one of the computed percentages of block mismatch is outside of a threshold level of standard deviations of the Gaussian distribution for the computed percentages of block mismatch; and

an identifier structured to indicate the video as having a scene change when both a difference metric outlier exists and a block mismatch outlier exists.

2. The video detector for detecting scene changes according to claim 1 in which the difference metric calculator calculates a root mean squared error between the pairs of adjacent video frames.

3. The video detector for detecting scene changes according to claim 2 in which the plurality of adjacent video frames are represented in YUV format and in which the difference metric calculator calculates a root mean squared error for a luminance component of the pairs of adjacent video frames.

4. The video detector for detecting scene changes according to claim 1 , further comprising an outlier detector verifier structured to verify that the difference metric outlier and the block mismatch outlier are based on data from the same pair of adjacent video frames.

5. The video detector for detecting scene changes according to claim 1 , in which the set of corresponding blocks of the second of the pairs of adjacent frames is configurable.

6. A method for detecting scene changes in an video that may include a number of different scenes, the method comprising:

accepting, at a video input, a video having a series of adjacent video frames;

dividing each video frame into a plurality of two-dimensional blocks;

computing a difference metric between each pair of adjacent video frames in the series of adjacent video frames;

identifying a Gaussian distribution for the computed difference metrics;

detecting whether a difference metric outlier exists, the difference metric outlier existing when one of the computed difference metrics is outside of a threshold level of standard deviations of the Gaussian distribution for the computed difference metrics;

comparing each of the plurality of blocks of a first of each pair of adjacent frames to a set of corresponding blocks of the plurality of blocks of a second of each pair of adjacent frames;

computing, for each pair of adjacent frames, a percentage of block mismatch based on the comparing and a total number of blocks in the first of each pair of adjacent frames;

identifying a Gaussian distribution for the computed percentages of block mismatch;

detecting whether a block mismatch outlier exists, the block mismatch outlier existing when one of the computed percentages of block mismatch is outside of a threshold level of standard deviations of the Gaussian distribution for the computed percentages of block mismatch; and

indicating the video as having a scene change when both a difference metric outlier exists and a block mismatch outlier exists.

7. The method for detecting scene changes according to claim 6 , in which computing a difference metric between each pair of adjacent video frames comprises calculating a root mean squared error between each pair of adjacent video frames.

8. The method for detecting scene changes according to claim 6 in which the series of adjacent video frames is represented in YUV format and in which computing a difference metric between each pair of adjacent video frames comprises calculating a root mean squared error for a luminance component of each pair of adjacent video frames.

9. The method for detecting scene changes according to claim 6 , further comprising, before indicating the video as having a scene change, verifying that the difference metric outlier and the block mismatch outlier are based on data from the same pair of adjacent video frames.

10. The method for detecting scene changes according to claim 6 , in which the set of corresponding blocks of the second of each pair of adjacent frames is configurable.