INTRODUCTION TO CONTENT-BASED VISUAL PROCESSING - Video Object Extraction and Representation: Theory and Applications - page 9

Figure 1.3. Image Segmentation: image segmentation problems often require a con-
textual knowledge to create a coherent segmentation: for instance, a) is there an
embedded triangle or a coincidental placement of semi-circles? b) could you find the
dalmation in the picture, if you didn't know there was a dalmation?

Figure 1.4. Motion Estimation As Source of Object Membership Information: a)
a frame of a coastguard sequence, a non-trivial image segmentation problem b) the
dense motion field from motion estimation techniques, revealing the moving coast-
guard ship from the background

temporal correlation. Instead of high-level image understanding such as
[Binford, 1982], we will depend on motion, a information source unavail-
able to image segmentation, for our high-level analyses such as object
detection and localization.

MOTION ESTIMATION
Motion estimation not only provides insight into object membership
(see Figure 1.4), but also provides valuable algorithmic techniques. While
image segmentation techniques are derived from spatial correlation, mo-
tion estimation extracts an another set of features derived from spatio-
temporal correlation [Lee and Blake, 1999]. Motion estimation derives