P1

why ?
Too many images, how to search ?
a. human annotation impractical
b. Content Based Image Retrieval - cbir identifies low-level features (color/shape/texture),
   need higher level because SEMANTIC GAP
   SEMANTIC GAP= (gap b/w lower-level features & semantic concepts used by humans to describe images)
c. Automatic Image Annotation

main idea of AIA :
automatically learn "semantic concept models" from large number of image samples
use the concept models to annotate new images.
=> images can be retrieved by keywords.

Img => Semantic Concept Model => Annotate.

P2 FE(Feature Extraction Review)
input image == unstructured array of pixels
1. FE from these pixels
2. region based FE >> global FE
3. region based FE needs Image Segmentation

Image Segmentation = want homogenous regions