Orientation invariant feature detection system and method for unstructured low-quality video

1. A method of detecting a human feature from a video, the method comprising:

obtaining a first video frame of images and a second video frame of images, the second video frame of images being later in time than the first video frame of images;

performing a luminance conversion on the first video frame of images and the second video frame of images;

performing a motion filter operation on the converted first and second frames of images;

performing a chrominance normalization on the second video frame of images;

performing a color filter operation on the normalized second video frame of images; and

processing the first and second frames of video images after the motion filter and color filter operations.

2. The method of claim 1 , wherein processing the first and second frames of video images after the motion filter and color filter operations includes morphological filtering.

3. The method of claim 2 , further comprising performing an AND operation of the morphological filtered first frame of video images and the morphological filtered second frame of video images.

4. The method of claim 1 , wherein performing a chrominance normalization includes removing brightness from the second frame of video images.

5. The method of claim 1 , wherein performing a luminance conversion on the first video frame of images and the second video frame of images includes examining brightness and not color in the first and second frames of video images.

6. The method of claim 1 , wherein processing the first and second frames of video images after the motion filter and color filter operations includes applying heuristic rules.

7. The method of claim 6 , wherein the heuristic rules are applied to a combination of the first and second frames of video images.

8. The method of claim 7 , wherein heuristic rules include conditions typical to faces in a teleconferencing scene.

9. The method of claim 1 , wherein the first and second frames of video images are frames from low-quality video.

10. The method of claim 1 , wherein low-quality video is video that contains features with less than 60×60 pixel resolution.

11. A system operable to detect feature images in video, the system comprising:

a central processing unit (CPU); and

a storage device coupled to the CPU and having stored there information for configuring the CPU to:

obtain video frames of images;

perform a luminance conversion on a first and a second frame of the video frames of images, the second frame being later in time than the first frame;

perform a motion filter operation on the converted first and second frames;

perform a chrominance normalization on the second frame;

perform a color filter operation on the normalized second frame; and

process the first and second frames after the motion filter and color filter operations.

12. The system of claim 11 , wherein the CPU is configured to communicate over a network.

13. The system of claim 12 , wherein the network is the Internet.

14. The system of claim 11 , wherein the storage device further includes information for configuring the CPU to morphologically filter the first frame after motion filtering with the second frame and the second frame after color filtering.

15. The system of claim 14 , wherein morphologically filtering includes morphological cleaning, morphological labeling, and morphological closing.

16. The system of claim 11 , wherein multiple frames of images on a videoconference are used as the first and second frames of the video frames of images.

17. The system of claim 11 , wherein the color filter operation involves use of a general color model.

18. An orientation invariant feature detection system for unstructured low-quality video, the system comprising:

means for obtaining a first video frame of images and a second video frame of images, the second video frame of images being later in time than the first video frame of images;

means for performing a luminance conversion on the first video frame of images and the second video frame of images;

means for performing a motion filter operation on the converted first and second frames of images;

means for performing a chrominance normalization on the second video frame of images;

means for performing a color filter operation on the normalized second video frame of images; and

means for processing the first and second frames of video images after the motion filter and color filter operations.

19. The system of claim 18 , wherein means for performing a color filter operation on the normalized second video frame of images includes using a Gaussian model of skin color independent of a specific person.

20. The system of claim 18 , wherein means for processing the first and second frames of video images after the motion filter and color filter operations includes applying heuristic rules.