). | Home | Table of Contents |
General Intelligence :
An illustrated example ("cube") is here (incomplete ).
The following is a very tentative outline, the order of steps may be wrong and additional steps may be needed...
1. Edge detection — distinguish between single-pixel lines and thick lines.
2. Classification of 1D lines: straight lines, curves, junctions. From this point on the representation is a collection of discrete lines and junctions: qualitative / structural rather than quantitative / spatial / continuous.
3. Good continuation — line / curve completion.
4. Contour simplification — the image is represented by both simple and detailed contours (allowing redundency in representation).
Some things can be recognized at this stage: text, handwriting, 1D graphics.
5. Characterization of regions according to color, shading or textures (segmentation).
6. Classification of 2D shapes
7. Extract gestalt-induced contours — a) numerous small, identical / similar elements may induce contours; b) the absence of features induces "invisible" contours
8. Deal with occlusion.
Some 2D graphics may be recognized at this stage.
9. Classification of 3D lines / surfaces.
10. Identify objects according to templates or algorithms. The question is how to define an object such as the object classes "bottle", "books", "faces", etc.
| Home | Table of Contents |
22/Mar/2006 (C) General intelligence corporation limited