Computer vision, image processing and computer graphics are 3 related domains, distinguished by their input and output.
- Computer vision
Input is an image, output is a description of the image. - Image processing
Input is an image, output is an (enhanced) image. - Computer graphics
Input is a description, output is a generated image.
Processing Pipeline
An image passes through 5 stages between the real world and its final reproduction.
- Acquire
Convert incident light, or a non-visible source (X-ray, UV, IR, sonar, MRI, radar), into a physical or logical representation. - Enhance
Correct defects introduced by the acquisition device: brightness, contrast, color consistency, geometric distortion. - Store or transmit
Encode the image for storage or transmission. Format conversion, compression, transmission errors and physical degradation (fading, scratches) can all introduce losses here. - Restore or recover
Identify and remove artefacts, restore brightness and color consistency, correct geometric distortion. - Reproduce
Convert image data back into an optical image using a printer, projector or display. Output depends on the device’s ability to reproduce data as light.
Light
Light is part of the electromagnetic spectrum, according to modern physics.
- Shares wave properties with X-rays, microwaves and radio waves, including dispersion
- Visible light spans wavelengths from 400 nm to 700 nm
- Red has the longest visible wavelength and the lowest visible-light frequency
- Violet has the shortest visible wavelength and the highest visible-light frequency
- Quantum physics attributes particle properties to light. Refraction is better explained by this theory.
Computer Vision Pipeline
Computer vision converts image measurements into a description of a scene or object.
- Acquire
Capture an image with a camera or another sensing device. - Pre-process
Correct noise, illumination variation and geometric distortion. - Segment
Separate pixels or regions relevant to the task. - Represent and describe
Extract features from the selected regions. - Recognize
Classify, locate or measure the objects from their features.
Applications
- Autonomous vehicles
Detect lanes, traffic signs, pedestrians and other vehicles to plan a safe path. - Medical image analysis
Detect or segment tumours, fractures and other abnormalities in scans. - Industrial inspection
Reject products with incorrect dimensions, missing parts or surface defects. - Security and surveillance
Detect, track and count people or vehicles in a monitored area. - Agriculture
Grade fruit, identify diseased crops and estimate yield from images. - Digital photography and videography
Improve captured images and apply visual effects.
