Carlo Tomasi, Computer Science. Human Vision Computer Vision?

Carlo Tomasi, Computer Science

Human Vision

Computer Vision?

Human Vision

?

Seeing is Interpretation

Seeing is Interpretation

Seeing is Interpretation

Seeing is Recognition

Giuseppe Arcimboldo, 1526-1593

Computer Vision

frog, mushroomreconstructionrecognition

Reconstruction: Stereo

Triangulation is Easy

Correspondence is Hard

?

It Can be Done (More or Less)

Face Recognition

Recognition: Face Space

Recognition: Face Space

Hardware HDTV video camera:

10M pixels, uniform 30-60 frames per second 20 Mbits per second (compressed) Variable field of view up to 1/3 of a sphere

Human eye: 7M cones in the fovea, 120M

rods 1.2M axons in the optic nerve 0.6Mbits per second

(compressed) 1/3 of a sphere field of view 28 arcsec resolution (finger at

30 m)

Hardware $1M buys you this from Dell:

1012 operations per second 1012 bytes of memory 1015 bytes of disk space 1012 bytes per second of communication

[speed of light 3x108 m/s]

One human brain gets you this: 1012 neurons 1015 synapses (connections)

[speed of action potential 100-102 m/s]

Comparing Apples and Oranges David Marr, 1945-1980, Cambridge (UK), MIT

The levels: Computational: What Algorithmic: How Mechanical: Wherewith

Computational goals are the same Algorithms can be similar

Brains are essentially parallel Computers are essentially sequential

Mechanical substrate is different

Paying the Bills