Image-Based Rendering - Introduction to Computer Graphics - Lecture Slides, Slides of Computer Graphics

Download Image-Based Rendering - Introduction to Computer Graphics - Lecture Slides and more Slides Computer Graphics in PDF only on Docsity!

Image-Based Rendering (IBR)

Human Visual Perception

• The human eye has two types of detectors
  • Rods and cones
• Rods, basically, only detect luminance and are the dominant detector in low light
• Cones detect color
  • Three types of cones: red, green, and blue (more or less)

Human Visual Perception

• To humans (with normal vision), green appears much brighter than other colors

RGB Color Space

• We can think of RGB defining a color space - In this case, a 3-dimensional color space

CIELab Color Space

• Very perceptually linear
  • Great for measurements and user studies
• Derived from the CIE 1931 calibration space, shown here

HSV Color Space

• Hue, saturation, and value
  • Hue: color
  • Saturation: “vividness”
  • Value: brightness
• Popular in graphics applications

CMYK Color Space

• Why do we need subtractive color for printing? - Paper is already white (maximum value), so adding ink can only make the image darker
• Black (K) is separate
  • Because no combination of cyan, magenta, and yellow can generate a true black

Color Spaces Review

• These were just a sampling of possible color spaces
• There are equations to easily switch between spaces - However, some colors that are within the gamut of one space may not be in the gamut of another
• Consider what properties you need when choosing a color space

Capturing Greater Dynamic

Range

• To capture this greater dynamic range with digital cameras, we can capture multiple bracketed images - Bracketing means taking multiple pictures of the same scene with different camera settings - i.e. different exposure times or aperture sizes
• To capture it with computer graphics, can just do lighting calculations with more bits

HDR Image Generation

• We have a problem here
  • Does anyone see it?
    • These images have too much dynamic range to be drawn on our display!
• The process of fixing this is called tone mapping

Retinex

HDR

Gaussian

Blur

Subtract

Ln(Image)

High

Frequencies

Low Frequencies

Scale by

k

Add

Result

k < 1. Low Frequencies High Frequencies

Input Image

Attenuated Low Frequencies

Exp(Imag

e)

How can we do even better?

• Maybe Gaussian filters aren’t the best tool
  • Blur across edges, obscuring high frequency detail
• Can use an edge-preserving filter
  • I won’t go into the math
  • Basically, the filter can recognize when it encounters an edge, and not blur across it

What Can We

Do?

Solution #3: Bilateral Tone Mapping

HDR Review Over

• Any questions?