Dr. Jeffrey Huang, Assistant Professor

1
N420 / D008 Multimedia Project
DevelopmentSpring 2001
Topic 2
2D Raster Graphics and Image Filtering

• Dr. Jeffrey Huang, Assistant Professor
• The School of Informatics, New Media Program,
  IUPUI
• e-mail huang_at_cs.iupui.edu

2
Digital Imaging

• Why digital?
• Analog vs. Digital
• Continuous vs. Discrete
• Ex 2D continuous signal to discrete
• Analog to Digital
• Sampling analog image are digitized by reducing
  their continuously varying value and color to
  discrete based on measurements or samples
• Sampling rate frequency
• Higher sampling rate higher resolution of image
• Resolution and Fidelity

0
0
0
1
1
1
Digital v.s, Analog
3 bit resolution (8 levels)
2 bit resolution (4 levels)
2D signal (image) example
Analog
128x 128 resolution
¼ of H and V sampling rate
3
The Pixel

• Image Pixel
• Short of Picture (pix) Element (el)
• Each of an images sample locations, together
  with its sampled value
• For color images, three samples are usually used
  for each image pixel R, G, and B
• Computer images based on pixels are thus
  sometimes referred to as sample-based-graphics.
• Sample-based graphics (raster graphics)
• Image pixel information is stored in a
  rectangular array and translated into screen or
  printer images in series of horizontal rows
  (called raster graphics)

Screen Pixel Printer Pixel
Halftoning
4
The Pixel (cont.)

• Image digitization

5
Digital Touch

• Touch
• The artists experience of making a mark in the
  process of creation
• Local-touch
• A mark controlled by the hand in process
  involving the accumulation over time of
  relatively small marks to create a larger image
• Example brush tools
• Global-touch
• Change an entire image area simultaneously,
  performing operations such as scaling or rotating
  an area or replacing one color with another.
• Different from local-touch process of
  accumulating marks over time.
• Imaging processing
• Mapping contrast, brightness, tonal mapping
• Filtering
• Example blur, sharpen, edge detection, image
  enhancement
• Algorithm touch and special effect
• Mathematics of fractals
• Automated texture creation Texture exploration

6
Image Filtering

• Filtering
• Vocabulary of filtering comes from signal
  processing, which deals with electrical sound
  waves over time
• Highly mathematical in image filtering
• Changes a pixels value by taking into account
  not only its original brightness but also the
  brightness of its neighbors
• High-frequency area
• Edge - High contrast area
• Texture wave
• Low-frequency area
• Gradients and soft forms Low contrast area
• Example skin, cloud
• Low-pass filter
• Remove high frequencies
• Make an image look blurred
• High-pass filter
• Remove low frequencies
• Make image look sharper

L.F
H.F
7
Low-pass High-pass filters

• Convolution
• The process of sequentially averaging whole
  neighborhoods in order to get new pixel values
• Filter kernel (convolution kernel)
• Kernel Size (matrix size) 3x3, 5x5,
• Kernel Shape rectangular, circle, diamond,

image
3
3
3
1
2.6
3
kernel
8
Low-pass High-pass filters

• Low-pass filter
• A simple blurring filter (Averaging filter)
• Example

A (A A top A left A down B) / 5
B ( B 4 neighbors)/5
9

• Example 1. Considering an image and a 3x3
  kernel (filter) shown below, what is the closest
  value of the high-lighted pixel after the kernel
  is applied onto image? (a) 2 (b) 3 (c) 4 (d) 5.
• Example 2. Considering an image and a 3x3
  kernel (filter) shown below, what is the new
  value of the high-lighted pixel after the kernel
  is applied onto image?

10
Low-pass High-pass filters (cont.)

• A less simple blurring filter
• Using weighted average
• Ex Gaussian Filter filter shape like a bell
• Average weighted according to different shape of
  kernels
• Gaussian filter
• The greater the filter radius or support, the
  more effect will be
• Sharpening filter
• Edge Detection
• Ex Fig. 2.22 Fig. 2.23

Filter radius
11
Transformation

• Translation
• Rotation
• Scale non-uniform scaling
• Flipped
• Distorted
• Filling the holes and anti-aliasing after
  transformation
• Nearest neighbor artifact
• Cubic interpolation
• Fit sin(x)/x surface
• Bi-linear interpolation
• 16 neighbors

12
Image Size

• File size
• Resolution
• Color depth
• Dimension
• Number of pixels in image
• height x width x resolution (no. of
  pixels/in2)
• Color depth
• 21 22 23 24 28 224 (True color)
• File size of pixels in image x color depth
•     Example 1. Given a color image with size
  of width x length 0.8 (in.) x 1.2 (in.) and
  the resolution of 7200 pixels/in2, what will be
  the approximate file size using true color
  without applying any compression (raw image)? (a)
  7K (b) 10K (b) 21K (d) 25K bytes.
• Example 2. Assuming a same image is given as
  above, what will be the minimum file size after
  converting this image into 16 gray levels without
  any compression applied? (a) 3.5K (b) 7K (c) 14K
  (d) 1.7K.

13
Image Resizing

• Zoom in and out
• Example of using zoom in / out
• 5x5 image on 300 ppi (pixels per inch) screen
• 20x20 on 72 ppi screen (in Adobe Photoshop)
• Example An 1¼ x 1¼ image, displayed on the
  computer screen of 300ppi, can be as large as
  (a)2.5x2.5 (b) 5x5 (c) 10x10 (d) 20x20,
  if it is displayed in Adobe Photoshop with the
  screen resolution of 72 ppi.
• Scale
• Re-sampling
• Need the filters (Fig. 2.36)
• Constant filters
• Triangular filters
• Gaussian filters

14
? - Channel

• Transparency Channel
• Let artist vary opacity for aesthetic effects and
  combine images easily in 2D/3D animation and
  video
• Forth Channel
• 4th frame buffer
• Store information how much of each color should
  contribute to the final image (how transparent)
• 0.5 50 transparent (50 colored)
• 1 untouched (remained at 100 of color value)
• Lighter values increasingly transparent
• Composite or merge images with varying levels of
  transparency