CSC 520

Digital Image Processing
Test 2

Name:______________________________ Date:__________________________

Instructions: Each student should perform there own work. Any collaboration by students will result in a grade of 0 for all students involved. If a question is unclear the student should list very careful there understanding of the question prior to answering the question.

1. Define isotropic in the context of an isotropic mask (kernel).

2. Write down the 3x3 mask for the vertical and horizontal Sobel operator.

3. Edge detection often results in breaks in the edges due to discontinuities. These breaks can be mended by edge linking techniques. In this problem the student will compute an edge image using the vertical and horizontal Sobel. The student will then implement edge linking via local processing as outlined in class where k = 10. (Reference: pg 726-727 Gonzalez/Woods 3rd.) The working image can be downloaded here (truck_rear.pgm).

The student will attach the following images to their test: 1) original image, 2) vertical edge image (mag), 3) horizontal edge image (mag), 4) vertical edge linked image (mag), 5) horizontal edge linked image (mag), and 6) logical OR of 4 and 5.

4. Image segmentation is a very important image processing step. The student will segment an image using the "Global Thresholding" method discussed in class. (Reference: pg 742 Gonzalez/Woods 3rd.) Using a stopping criteria, delta T = 5 pixels and the starting T = 200. The image for segmenting can be downloaded here (fingerprint.pgm).

The student will provide a plot of the histogram of the original image, the threshold value T for segmentation (label T on the histogram plot), and the segmented image. NOTE: a segmented image will be binary where black is 0 and white is 255 for an 8-bit image.

5. Often images contain some noise components that are related to either the sensor or the environment. The actual image is a combination of the ideal image (no noise) and a noise component construct from a PDF. If the noise component is understood it can be removed by a restoration technique. Determine the type of noise that effects the following image (noisey_image.pgm) by evaluating the PDF of a sub image. (See figures 5.4 and 5.6 for guidance.)

The student will select an appropriate sub image such that the histogram of the sub image will provide a functional PDF from which the student can determine the type of noise (uniform, salt-pepper, Gaussian, gamma (Erlang), or exponential. The student will compute the PDF parameters for the noise from the average (mean) z-bar and variance. NOTE: that if the noise is modeled as Gaussian then the only parameters required are mean (z-bar) and variance.

6. The student will implement two advanced spatial filters on the noise image from 5: 1) Geometric mean filter and the alpha-trimmed mean filter. The sub image dimension (m,n) for each filter is 5,5.

The student will compare the results of the two filters in a succinct discussion. The outputs are: 1) original image, 2) the geometric mean result and 3) the alpha-trimmed mean result.