%----------------------------------------------------------------------------

VISION - Low-Level

%----------------------------------------------------------------------------

The vision task:

Input: 	A 2-dimensionsal image (typically, a bitmap)
Output: Recognition of objects that produced it 
        (including their shapes, positions, colors, sizes, etc.) 

Both visual recognition and language understanding are examples
of abductive inference, i.e., inference from effects to causes.

%----------------------------------------------------------------------------

Why is vision hard?

Computational complexity

Ambiguity 

For example, the image:

                ---------
                |       |
                ---------


may be produced by any of the following objects:

           /--------/|     ---------       /-------/\
          /        / |     |       |      /       /  \
         /--------/  /     |       |     /-------/----\
         |        | /      |       |
         ----------/       ---------



%----------------------------------------------------------------------------

Stages in vision processing (Marr):

Raw 2-D              Primal              2-1/2 D                3-D
Image    ----------> Sketch  ----------> Image      ----------> Image
(pixels)             (edges)             (surfaces)             (volumes)

         low-level	     intermediate           high-level
	 processing          processing             processing



2-1/2 D Images are also called intrinsic images

Control of Processing:
	
	mostly bottom-up in low-level vision
	both bottom-up and top-down in high-level vision:
      		model-driven vision

%----------------------------------------------------------------------------

LOW LEVEL VISION:

    Example: EDGE DETECTION
    Assumption: Gray image (no color)

        What is an "edge"?
        How can we detect an "edge" in a pixel array?

        By comparing the intensities of neighbouring pixels.

Example:

	If the intesities were

	9 8 3 2 2 3
	9 8 3 2 3 4
	9 9 4 3 3 2
	9 7 3 4 2 2
	9 8 3 3 2 3

	then an edge between the second and third columns is possible
	(because of the ``large'' difference in the intensities).

%----------------------------------------------------------------------------

Copyright (c) Ashok Goel, 1994-97
Associate Professor, College of Computing
Georgia Institute of Technology, Atlanta, Georgia 30332
Email: goel@cc.gatech.edu

%----------------------------------------------------------------------------