Vision Science Quiz

Computer Vision

• Lecture presented by Dr. Ahmed Taha, Computer Science Department, Faculty of Computers & Artificial Intelligence, Benha University
• Topic covers human vision, low-level vision, mid-level vision and high-level vision.

Human Vision

• Eyes are the sensors for vision
• The brain processes visual information.

Low-Level Vision

• Photo manipulation including size, color and exposure
• Feature extraction including edges and oriented gradients, segments using X-Pro II
• Key aspects of visual processing

Mid-Level Vision

• Image relationships to image and world
• Methods like panoramas, multi-view stereo, structure from motion, structured light and LIDAR (Light Detection and Ranging)
• Connections between images and time using techniques like optical flow and time-lapse

High-Level Vision

• Image classification and relationship to semantics
• Object detection and segmentation
• Applications like retrieval and robotics

Is Vision Easy or Hard for Humans?

• A discussion prompt.

Eyes - The Sensors for Vision

• Eyes are the receptors for visual stimuli.

Why do things have eyes?

• Visual stimulus is an important signal
• Photoreceptive proteins (eyespots) were the starting point for vision

Eyespots - the beginning of vision

• Eyespots are photosensitive proteins without surrounding structures.
• They are sensitive to ambient light.
• Euglena swim toward light for better photosynthesis.
• Snails move away from light.

Pit eyes - the first eyes?

• Photosensitive cells in pits.
• Block some light to get better direction information about light.
• Very common in animal phyla (28 of 33).
• Simple structure with low acuity.

Simple -> Complex Eyes

• The evolution from simple to complex eyes.
• Different types of simple and complex eyes are described (mirror, pinhole, lens, multi-lens, refractive cornea, homogeneous vs inhomogeneous)

Eyespots - no acuity

• Eyespots lack acuity.

Pit eyes - some acuity

• Pit eyes have some acuity.

Complex eyes - high acuity

• Complex eyes have high acuity.

Refraction: more light + acuity!

• Refraction increases light and improves acuity in eyes.

Focussing: changing refraction

• The eye changes its refraction to focus on near and distant objects.

Complex eyes - a huge advantage

• Many different eye mechanisms account for many ways cameras work today.
• Better visual acuity is a common goal.
• More than 96% of known species have eyes with this function.

So how do human eyes work?

• Light passes through the cornea, humors, and lens to focus.
• Light is detected by photosensitive cells on the retina.
• Information is transmitted to the visual cortex via the optic nerve for processing.

Human Vision (Diagram)

• Diagram illustrating the structure of the human eye and the path of light.

How do we process this light?

• Photoreceptive cells (rods and cones) detect light. (~120 million in the retina)
• These cells aren't evenly distributed, nor are they all the same. Rod cells and cone cells each differ in sensitivity.

Fovea: where it's all happening

• The fovea is a small region of the retina where cones are densely packed (approximately 200,000 cones/mm²)
• Fovea has highest visual acuity, allowing for highly detailed vision and reading.

Peripheral vision: don't get eaten

• Peripheral vision detects a wide field with fewer cones and more rods.
• This allows for seeing movement and broad shapes outside the central focus.

Entire Visual Field in One Eye

• A diagram about the field of vision in one eye.

Photoreceptors need change!

• Photoreceptors must adjust to changing light conditions to retain sensitivity.

Fixational eye movement

• Eye movements to maintain optimal perception (microsaccades, ocular drift, microtremors).

The Brain - A Visual Processor

• The brain is crucial for processing visual information.

Brains (maybe) came after eyes!

• Eyes connected directly to muscles; Some jellyfish don't have brains.
• Brains and eyes coevolve: more complex eyes lead to more complex brains.

Ganglia transmit info to brain

• Ganglia (approximately 1 million) process data from photoreceptors.
• M cells respond to motion, depth and orientation.
• P cells focus on color, shapes and fine details.

Visual cortex interprets data

• V1: primary visual cortex for edge detection and high spatial sensitivity
• V2: secondary visual cortex for color, shape, size and possibly short term memory. Signals move on to V3, V4, V5.

Visual cortex is split (maybe)

• Ventral stream for recognition and object details. Dorsal stream for action, movement, and spatial relations

Ventral vs dorsal stream

• Comparison of ventral and dorsal stream characteristics (function, sensitivity, memory, speed, consciousness, frame of reference, visual input, monocular vision).

What does this split mean?

• Recognition and action are separate systems (damage to the dorsal system affects visual control; damage to ventral affects recognition).

Blindsight: vision without sight

• Phenomenon where people with damage to visual processing areas can still respond to visual stimuli unconsciously.

The brain and vision

• The visual cortex is a major part of the brain.
• Roughly 30% of cerebral cortex and half of electrical activity is devoted to vision.

Case study: How the brain sees 3d

• Perspective cues like focus, blur, parallax are used for 3D perception
• Two eyes give slightly different images for stereo vision; eyes converge to depth.
• Brain interprets kinetic depth, occlusion, familiar objects, and shading for 3D vision.

So what are we looking at anyway?

• Visible light is a range of electromagnetic radiation (~400-700 nm).
• Light's properties as a wave or particle are discussed.

Visible light and the sun

• The spectrum of the sun.

Light is a combination of waves

• Light is a composite of multiple waves with varying amplitudes.

Sources of light are diverse!

• Different sources of light (ruby laser, Gallium Phosphide crystal, Tungsten lightbulb, normal daylight) have different spectral distributions.

"White" light - all wavelengths

• White light contains all wavelengths of the visible light spectrum.

Objects reflect only some light

• Objects absorb or reflect the varying wavelengths of incident light.

What color is the object?

• Object color depends on the incident light and reflectivity of the object.

Different illumination matters!

• Illumination affects color perception.
• Different light sources (daylight, sunset, fluorescent, LED, incandescent, cool fluorescent) have different spectral power distributions.

Case study: makeup application

• Makeup effect on skin tone depends on the spectral distribution of illumination light.

Photoreceptors and light

• Photoreceptors (rods and three types of cones) have responsiveness curves.
• Rods are sensitive to a broader range of light while cones are sensitive to specific wavelengths, with short, medium, and long cones.

Photoreceptors and light (Diagram)

• Graph showing sensitivity curves for different photoreceptors.

Cones and color

• Our color perception results from different cone responses to various wavelengths.
• Each cone's output curve is integrated to determine the perceived color.

All cones are not equal

• Cones are not uniformly distributed in the visual field.

Different wavelengths are brighter

• Graph displaying brightness of various wavelengths perceived by the eye.

Color is our perception of waves

• Different waveforms can look the same (metamers).
• Our perception (color) differs from the actual wave form itself

Metamers are great!

• Metamers' create various colors with just three outputs in RGB systems.

CIE 1931 and Color Matching

• Color matching experiments using color controls to produce a certain color
• Experiments were conducted by scientists Wright and Guild using color controls and primary lights.

Now we can make a map of color

• Color space map demonstrates how different wavelengths create colors and how they are different from each other

Linear colorspace

• Mathematical description of how color is related.

"Theoretical" CIE RGB primaries

• Theoretical primaries for color creation

Practical sRGB primaries, MSFT 1996

• Practical primary colors for color creation

MANY different colorspaces

• Multiple different color spaces for color creation

What does this mean for computers?

• RGB is a system to create various colors with just three outputs for use in computers.
• Humans differ in how they perceive colors.

Grayscale - making color images not

• Grayscale images are simulations of monochromatic images from RGB where color information is not included
• Gamma compression is used to represent information for brightness

RGB is a cube...

• A visual representation of the RGB color space as a cube.

Hue, Saturation, Value: cylinder!

• A model of color perception based on hue, saturation, and value. A representation as a cylinder is displayed.

Hue, Saturation, Value

• Light perception model based on hue, saturation, and value.

Thank You

• Presentation closing.