Video Segmentation
- Basic Concept: Shot
Shot(take): contiguous recording of one or more video
frames depicting a continuous action in time and space.
Boudaries between consecutive camera shots -> video
segments finding.
- When is it easy? Abrupt
scene changes.
Simple method: examining frame-to-frame intensity
differences at the pixel level.
Shortage: camera movement, noise, and illumination
changes may contribute to intensity differences.
- Two categories of
segmentation method: frame difference techniques and
histogram
Frame Difference Techniques:
Dx,y(t) = (||Sx,y(t) - Sx,y(t+1)||), for any x, y belonging to S
- Improvement method: Utilizing the number
of pixels undergoing changes from one frame to the next
by taking the difference of sub-sampled images and
thresholding at a suitable value. Advantage of this
method: it can cope with noise, minor illumination
changes, and camera motion.
- Histogram Techniques
Interframe difference is susceptible to noise,
illumination changes and camera motion. Histogram of a
video frame provides a global description about the
apperance of an image to avoid this sensitivity.
How to compute a histogram: Measuring probability
distribution of pixel values in the entire image.
We can also combine the knowledge
of frame difference and histogram to identify different
cut types. Approach is based on modeling of the
distribution of pixel difference.