Rolling Dev's blog

Computer Vision: 2. Perception

yoonmin's photo
yoonmin
·

3 min read

Table of contents

2.1 Properties of Light

Light

  • is an electromagnetic radiation (EMR)
  • λ: wavelength

Plenoptic Function

R(X, Y, Z, θ, Φ, λ, t)

  • 3 position, 2 angle, strength, time

The Light Field

R(u, v, s, t)

  • where light captured between two plane (u, v) and (s, t)

The Visible Light Spectrum

  • has wavelength between 380 to 720nm

Spectral Power Distribution (SPD)

  • means how much the light source has light power relatively to wavelength

2.2 The Human Visual System

2.2.1 Color Perception

  • light -> retina -> photosensitive cells

    Density of Rods and Cones

    image.png
  • are not uniformly distributed on the retina
    • rods: responsible for intensity
    • cones: responsible for color
    • both adapt to be more sensitive in low light

      Visual Acuity

      image.png

2.2.2 Brightness Contrast

  • depends on the surrounding region

    Visual Dynamic Range

  • depends on the luminance level

    Light Response is Nonlinear

  • gamma correction
    • actual linear brightness != perceptual linear brightness (sRGB)

2.2.3 Fixational Eye Movement

  • 3 types
    • Microsaccades: short and fast linear mov.
    • Ocular drift: constant slow mov.
    • Microtremors: tiny vibrations for seeing fine details

2.2.4 After Images

  • occurs when tired photoreceptors send out negative response after a strong stimulus

2.3 Color Perception

Three Types of Cones

  • S, M, L
    • Short (S): blue
    • Medium (M): green
    • Long (L): red
  • have different sensitivities: green > red

    Color Perception

  • filtered color = multiply its response curve by the spectrum, integrate over all wavelengths

    Metamers

2.4 Camera and Optics

2.4.1 Camera Response Function

  • eventually pixel values

    Recovering the camera response

  • Model every step
  • Calibrate (estimate) the response function
  • Calibrate the response function from several images

2.4.2 Image Formation

  • just film?

    Pinhole camera

  • pinhole barrier -> film
  • produces blurring
  • camera obscura (the first camera)

    Aperture

  • too small -> blurring
  • too big -> blurring as well
  • need bright environment or more sensitive film

    Adding a Lens

  • collect more rays
  • focal length; distance(center of lens, focal point)
  • depth of field
    • small aperture -> less rays, clear and deep background
  • circle of confusion

2.4.3 From Light to Pixels

  • IR-blocking filter
    • necessary in RGB sensor system
    • unnecessary in brightness sensor system
  • Color filters
    • sensors are color blind.
    • filtering colors to each sensor will be helpful.

Bayer filters

  • pixels of square grids
    • 1/4 of pixels -> see red light
    • 2/4 of pixels -> see green light
    • 1/4 of pixels -> see blue light

Debayering

  • is to interpolate missing values

RGB images (of three channel)

  • instead of bayer filters, we can take picture of each color and add together.
  • no missing RGB values
  • expensive

Images From

Computer Vision