ENGG 199: Digital Image Processing

Winter Term 2009

Course Description

Digital image processing has come into widespread use in many fields including Medicine, industrial process monitoring, military and security applications, as well as satellite observation of the earth. This course will cover many aspects of image processing which many graduate students will find valuable in their research. Topics will include: image sources (e.g. cameras, scanners, medical devices MRI/CT/Ultrasound, synthetic images), computer representation of images and formats (e.g. JPEG, RGB, TIFF, look up tables), operations on images (e.g. denoising, deblurring, geometric transformations, histogram equalization) and image analysis (e.g. segmentation, pattern recognition). In this course we will stretch the conventional notion of images from 2D pixel arrays to include 3D data sets and we will explore how one can process such stacks of voxels to produce useful information. This course will also touch on some advanced topics in image processing which find their applications in medical imaging such as segmentation, co-registration of images from multiple sources using mutual information, stereopsis, and coordinate transformation in 2D and 3D using affine transformation matrices. This course will require the completion of a project selected by the student.

Instructor

Instructor: Alexander Hartov
Location: Room 231, Cummings Hall
Telephone: 646-3936
E-mail: alexander.hartov@dartmouth.edu

Prerequisites

ENGS 92 (Fourier Transform and Complex Variables or equivalent), ENGS 103 (Statistical Methods in Engineering or equivalent).

Topics Covered

  • Review of Matlab
    • Matlab Basics: scripts vs functions, memory usage, vectorization.
    • Matrix operations
    • Image representation in Matlab: uint8 vs float, single plane vs 3-plane, flat vs LUT
    • Loading and displaying images
    • The Matlab Image Processing Toolbox
  • The Perception of Color by the visual system
    • Additive (RGB) vs subtractive (YCM) color representation
    • RGB encoding in television/video
  • Image sources
    • Camera technology: CCDs spatial and signal resolution, pixel array, color capture, multiple passes vs 3-CCD sensor arrangements
    • Scanners
    • Medical devices: MRI, CT, Ultrasound
    • Synthetic images
  • Computer representation of images and formats:
    • Color representation schemes: RGB, YCrCb, YUV
    • RGB , JPEG, TIFF, look up tables (LUTs)
  • Operations on images
    • Basic distinctions: Spatial frequency domain, point, and kernel based operations.
    • Image improvement: denoising, deblurring, histogram equalization
    • Geometric transformations: affine transformation in 2D and 3D, the homogeneous coordinates system.
    • Image analysis: Segmentation (pixel values based, watershed algorithm, level set algorithm, pulsed neural network algorithm), pattern recognition (pattern matching, neural network based recognition).
  • 3D Imaging Datasets:
    • Image stacks from MRI and CT sources
    • A brief overview of the DICOM format, Matlab commands to read DICOM data
    • MRI/CT coordinate systems, pixel/slice indexing and spatial dimensions, right-handedness of implicit coordinate systems
    • Spatial transformation of 3D data
    • Co-registration between multiple sources of imaging data using mutual information
  • Stereopsis:
    • Basic theory
    • The camera calibration problem
    • Image rectification
    • 3D data extraction

Grading

Grading will be based on:

  • Homework problems (one set per week): 40%
  • Project reports: 20%
  • Mid-Term examination: 20%
  • Final examination: 20%

Textbook(s):

No book purchase will be required. Extensive class notes will be distributed and will refer to (available at Feldberg Reserve or as PDFs on the BlackBoard server):

  • Image Processing Handbook, Russ
  • Biomedical signal and image processing, Najarian & Splinter
  • Digital Image processing using Matlab, Gonzalez, Woods & Eddins.
  • Image processing using pulse-coupled neural networks, Lindblad & Kinser.

In addition, papers on specific topics (e.g. stereopsis, watershed transform, pulse-coupled neural networks) will be assigned for reading.

Special Requirements

Most homework assignments will require the use of computers and Matlab.

Blackboard

More information about this course can be found at the Blackboard site. You can login to Blackboard using your DND username and password.