ECS 173: Image Processing & Analysis

Summary of Course Content

1. Analysis of 2D Images (photographs)
   1. Low-level information extraction: Frequency-domain representation of images, edge and corner detection, and texture analysis will be discussed in detail. Specific algorithms to be discussed include derivative of Gaussian edge detection, Harris corner detection, and texton estimation.
   2. Image features: Students will learn various constructions for extracting relevant features from images. Students will learn about filter bank approaches, geometric and illumination invariants, and linear subspace decompositions of image patches. Identification of salient surface patches will also be discussed.
   3. Object recognition: Students will receive an overview of methods for describing and detecting objects in 2D images. Appearance-based methods based on principal components analysis, convolutional image filters, and raw image classification will be described. Shape-based object detection based on constellations of object parts, local edge features, and alignment to prototype shapes will be presented.
2. Analysis of 3D surface imagery
   1. 3D surface parameterization and representation: Representation of 3D surfaces based on points, parametric surface models, patches, and geons will be discussed. The effects of noise, partial occlusion, and sensor artifacts on these surface descriptions will also be described.
   2. Automated model building from surface data: Semi-automated and fully-automated procedures for building complete 3D models from collections of partial 3D sensor data sets will be presented. Semi-automated techniques based on landmark placement will be discussed. Fully-automated techniques based on local surface descriptors, constrained data collection, and global surface descriptors will be shown as well.
   3. Object recognition in 3D data: Detection of objects in 3D surface data sets will be discussed. The effects of partial occlusion and clutter will be discussed. Techniques based on alignment, local surface descriptors, and machine learning approaches will be described.
3. Analysis of volumetric images
   1. Low-level processing: Students will learn approaches to correct for image artifacts found in computed tomography (CT), magnetic resonance (MR), and functional MR images. Bias field correction, blowout, and scattering artifacts will be discussed.
   2. Description and modeling of biological shapes: Methods for mathematically describing biological objects found in volumetric images will be presented. Representation of 3D solids will be discussed, and computational anatomy approaches to representation of populations of shapes will be presented.
   3. Localization and detection of objects: Principles and algorithms for localizing biological shapes in volumetric images will be presented. Shape-model-based techniques for estimating the location of constrained, expected objects such as the brain will be discussed. Low-level detection of amorphous, unexpected objects such as tumors and calcifications will also be presented.

Illustrative Reading
Terry S. Yoo (Editor), Insight into Images: Principles and Practice for Segmentation, Registration, and Image Analysis, AK Peters Publishers, July 29, 2004

Potential Course Overlap
Volumetric images are also treated in ECS 177, but the emphasis there is on the visualization rather than automated extraction of high-level data. Two-dimensional image processing is treated in greater mathematical depth in EEC 206 and EEC 208, graduate courses which are currently taught infrequently.