Feature Construction and Feature Integration in Visual Learning
We propose visual learning methods. Visual learning based recognition methods can be flexibly
applied to wide variety of recognition tasks because they can find useful features of objects
for recognition by learning with given images. In order to construct such recognition systems,
we propose two types of visual learning methods based on the appearance and the shape of objects.
In both methods, low-level features are combined into a higher-level feature. The appearance based
method combines some small parts of the contours of an object into a key shape to recognize the
object. The region based method combines some local features of an object into a global feature
which can fully represent the whole object. Since an object can be fully described by the appearance
and the region components, we propose the more powerful method by integrating these two methods.
[Back]
