Digital Signal Processing Reference
In-Depth Information
Figure 6.2.
The shape of object projection can aid classification and provide content.
For instance, in a), if we classify the object as a person, we can infer the height and
pose of the person. Shape is also naturally ambiguous and the same shape have very
different meanings such as b), where the picture shape can either be a two faces or a
candlestick.
easily and more reliably desired content than an automatic query
system. A user interface that allows manual guidance of similarity
measure can aid the accuracy of the query system.
4.
Speed
Computational cost of calculating the similarity measure should not
be a limiting factor: thousands of shapes should be compared in a few
minutes. The computational cost of extracting the shape descriptors
is a secondary concern, since the extraction is done off-line.
5.
Compactness
Shape descriptors should be on the size of the compressed raw bitmap
or smaller.
For our shape descriptor in particular, we also describe a new func-
tionality:
Motion Annotation
A descriptor that captures the temporal aspect of video information
provides users with more powerful and descriptive search criteria.
Since the shapes exist within the context of video sequence, motion
annotation of shape descriptors may be a powerful query tool.
3. SHAPE EXTRACTION VIA VOS
In this section, we describe how we extract shape information by using
assumptions about object content and derive a representation that is
Search WWH ::
Custom Search