- Binary Image Analysis
- Be able to explain how histogram-directed thresholding works
- Be able to show how the pixel-by-pixel classical connected components algorithm or the efficient run-length version works on a small sample image.
- Be able to use the operations of mathematical morphology (dilation, erosion, closing, and opening) to perform a specified task on a binary image. Be able to use box or disk structuring elements or to design your own structuring element as needed.
- Be familiar with the meanings of the geometric and shape properties of Chapter 3, so that you can apply them if needed to a classification problem.

- Filtering and Edge Operators
- Be able to apply mean, Gaussian, or median filters to a portion of a given image to produce an output image.
- Be able to understand and explain statements in the filtering notation of this lecture, where images can be thought of as functions or as 2D arrays.
- Be familiar with the definitions of cross-correlation and convolution and the difference between them.
- Be able to apply a simple differential edge operator such as the given Sobel or Prewitt operators to an image and to show the gradient vector, its magnitude, and its direction.
- Be able to explain the motivation for the zero-crossing operators
- Be able to give a reason why the Canny operator has been the most popular for multiple years, based on what it does and what it might be used for.

- Color and Texture
- Be able to use different color spaces as needed for problems
- Be able to use the Swain-Ballard color histogram distance to compare 2 images.
- Be able to answer questions about how K-means clustering works and to use it as a step to solve some larger problem.
- Be able to compute simple texture measures including edgeness per unit area, magnitude and direction histograms, local binary pattern histograms, co-occurence matrices, and features derived from them.
- Be able to answer questions about the Laws operator.
- Be able to explain the scale invariance of the Blob World texture features.

- Advanced Segmentation
- Be able to explain the difference between region growing, split-and-merge, and clustering as methods for segmentation.
- Be able to compare Ohlander's recursive histogram-based clustering to Shi's Graph Partition, which is also recursive.
- Be able to find normalized cuts in a given graph and use that to segment the graph.
- Be able to answer questions about the EM clustering algorithm including how it differs from or extends K-means.

- Lines and Arcs
- Be able to apply the Hough Transform for straight lines or for circles to a given piece of an image.
- Be able to show how the Burns line finder would work on a piece of an image.
- Be able to show how to find consistent line clusters, given the line segments and their properties.

- Content-Based Image Retrieval
- Be able to discuss the differences between the sample systems
discussed in class (take a look at them on the web, too):
- IBM's QBIC
- UC Berkeley's Blobworld
- UW's FIDS

- Be able to answer questions about or show how to use the
following image distance measures:
- color histogram
- gridded color
- texture histograms
- gridded texture
- shape projections
- tangent-angle histograms

- Be able to answer questions about Rowley's face finder.
- Be able to answer questions about Fleck and Forsyth's Flesh Detector.
- You don't need to study the JFS distance metric using wavelets.
- You don't need to study the section on relevance feedback.

- Be able to discuss the differences between the sample systems
discussed in class (take a look at them on the web, too):
- Pattern Recognition
- Be familiar with the terminology of classifiers
- Be able to answer questions about or use for classification
- Nearest mean classifier
- Nearest neighbor classifier
- Bayesian decision making
- Decision tree
- Simple neural net

- Be able to answer questions about the two different entropy-based
methods for constructing decision trees from training data.
- Information gain
- Information content