CSE5519 Advances in Computer Vision (Lecture 2)

A brief history (time) of computer vision

Theme changes

• Prominence of computational and software concerns (was always constraining, always unevenly distributed)
• Optimization - not procedural reasoning
• Data and learning - no rules
• Rich appearance and discriptors - no simple point and line features
• Deformable template for recognition
• In the long run, data and computation win over cleverness - bitter lesson .

The pre-historic goes further back

1960s: Image processing and patter recognition, blocks world

First OCR conference in 1962.

Bubble chamber photo.

Microscopy, cytology.

Face recognition.

Fingerprint recognition.

CVPR: Computer Vision and Pattern Recognition.

MIT centered narrative.

Representations extraction and descriptors.

1970s: Key recovery problem is defined

Structure from motion, stereo, shape from shading, color constancy. Attempts at knowledge-based recognition.

Recovery of shape from image data.

Syntatic/procedural recognition systems.

1980s: Fundamental and essential matrix

Multi-scale analysis, corner and edge detection, optical flow…

Structure from motion: RANSAC.

3D vision.