CE 261: Digital Image Processing, Review

This sheet is intended to serve as a list of concepts and material that are important to study for the comprehensive final that is worth (30%) of your grade. Material in homeworks is important as well, and is listed separately, below.

Week [1] of April 7:  Intro to image processing, (Ch. 1 & Ch. 2 & Ch. 3)

L1, goals for the class, general introduction: Low, medium, and high
level image processing. 2 papers that I presented just as intro to my
work.

L2, what is an image? Electromagnetic spectrum (ROYGBIV), discrete
I[x,y], continuous I(s,t). Quantization, reals to quantized levels.
High, mid, low level as knwoledge, features, pixel pushing.

 medical imaging example, NeoPath. Digitization.
  
Week [2] of April 14: Image processing applications, Image Vision Libraries, Adobe 
Photoshop, Biological Basis for vision, Color theory (Ch. 4, Ch. 21) HW #1 Lab #1

 L3,  elements of digitizing images, aperture, mechanism, sensor, A/D,
storage. Types of sensor. CCD-charge coupled device solid state cameras.
Full fram, interline, frame transfer.

Gaussian spot profile.  Monitor design tradeoffs, flat field,
line patterns, checkerboard.

 L4, software engineering, structured programming, OOP (object oriented
programming). design, test, documentation. ImageVision Lib. intro. File
formats, PPM as an example. Image model again. rods, cones, horizontal
cell, ganglionic ell, retina, optic nerve, chiasm, lateral genicualte 
nucleus, visual cortex. similar tirangles.

  
Week [3] of April 21:    Histogram/Point processing (Ch. 5 & Ch. 6 & Ch. 7), HW #2

 L6. Color theory, tristimulous color theory, RGB (red green blue),
HSI (hue saturation intensity), HSI to RGB conversions. point processing.
Transfer function, histogram,
contrast operations. histogram equalization. requantization.

 L7. affine transformations, geometric warping, bilinear, nearest
neighbor filtering. backwards, forwards, affine combination,
efficiency.

Week [4] of April 28:    Geometric transforms and image warping, permutation warping 
(Ch. 8 and handouts) HW #3 Lab #2



L8 Wolberg, 2 pass mesh warping algorithm. permutation warping.

Week [5] of May 5:    Image based rendering, Light Field/Lumigraph, View morphing, 
three dimensional image processing ((Ch. 8 review )Ch. 22 part, handouts) Test 1. 

 L9 image based rendering and modelling.

 L10 midterm (important stuff).
  
Week [6] of May 12:    Image based rendering continued, multibase plane stereo/epipolar 
geometry, reconstruction of camera intrinsic and extrinsic parameters, voxel coloring. 
(Ch 15 part, handouts) HW #4

L11. Stereo, boresighted camera geometry, disparity, range, matching.
IBR survey of techniques, silhouettes, contours, lines, stereo, 

 L12. review stereo. IBR techniques, view interpolation,
light fields, lumigraph, plenoptic rendering, environment map, continuum
of techniques.

Week [7] of May 19:   Introduction to linear systems theory and The Fourier transform 
(Ch. 9 & Ch. 10) HW #5

 L13. impulse response, Fourier transform , transfer function,
convolution theorem. spatial temporal domain, frequency domain. 
Linear systems theory.

 L14. Guest lecturer Glen Langdon, Fourier transform, and
Discrete Cosine transform. Fourier series, 4 point/8 point DCT,
JPEG, properties of Fourier transform, shift invariance, 
linearity (again), harmonic analysis.

  
Week [8] of May 28:   Sampling theory and discrete transforms (Ch. 12 and Ch. 13) 
HW #6 Lab #3 May 26 Exchange day, No class on Tuesday !)

 L15. Sampling theory, frequency, period, Nyquist's limit,
superposition, time invariant, convolution in discrete domain,
ideal reconstruction, higher frequency means wider spacing of
frequency replicants, aliasing, sinc ftn. comb ftn.

  
Week [9] of June 2:    Low level image processing: Image restoration (Ch. 16) HW #7

 L16. KLT example application, mean square error, eigenvalues
eigen vectors, covariance matrices. basis functions, domains, spaces.
GOES satellite multispectral data, image segmentation, image compression,
DCT as approximation to optimal basis (KLT). Generic formulation of
discrete image transforms t[u,v] = SUM SUM f[x,y] g[x,y,u,v] where
g[] is the kernel, and f[x,y] is input. kernel different for different
transforms.

 L17. Restoration, deblurring, system modelling, noise modelling.
model, spatially invariant system with noise. Wiener deconvolution,
power spectral density. Visual system doesn't care about mean square
error. power spectrum equlization, geometric mean filters, linear
algebraic restoration. 
  
Week [10] of June 9:   Mid level image processing: Segmentation (Ch. 18) Lab #4

 L18, pseudo code, segmentation, thresholding, optimal thresholding,
gradient operators, sobel, laplacian (double derivative), zero crossing,
noise sensitivity of gradient, 4-connected, 8-connected, connected
components. area defined by histogram. transitive closure, edge linking,
median filtering. 

 L19.  Hough transform, mathematical morphology, opening,
closing, dilation, erosion, object membership map, boundary chain
code, run length coding.

 Homeworks: 
 1 ,  number of combinations possible with a base, and
a given number of digits is (base) to the power of (digits),
Charge storage density. dynamic range, dark current, noise.
modulation factor M = (D(0,0) - D(1,0))/D(0,0).
 2,  histogram, contrast enhancement, area of pixels.transfer
function. histogram calculations. Color lookup tables.
 3.  interpolation. bilinear interpolation. transforms.
translation, rotation. scaling. 
 4.  stereo display on a monitor. solving for true range.
detecting features in intensities. disparity. 
 5.  Fourier transform. superposition. induction.
convolution. box function. triangular function. testing
for linearity. integration by parts.
 6.  shah function or impulse train or comb function.
Nyquist's theorem/critical sampling limit fs > = 2fb
fs - sampling frequency, fb - bandlimit frequency in 
sampled image. impulse response. convolution in discrete domain.
 7. edge spread function. line spread function. Modulation
Transfer Function (MTF), optical transfer function (OTF).
Discrete Fourier Transform. decomposed kernels by SVD. 
area implied by number of pixels in histogram. area of
projected sphere, volume of sphere 4/3 pi r ^ 3.
Area = integral { T to infinity} of H(D) dD.


craig_wittenbrink@hpl.hp.com
Last modified Thursday, 11-Jun-1998 14:17:12 PDT.