ECE 595 Computer Vision Assignment 3

Submit Matlab code, images, and a Word document describing your work.

Download images for this assignment from cv03.zip.

  1. Use an image showing bright colors as the test image for components.m. Discuss what information the different components convey.

  2. Design a 3-D line illustration of the third-floor KL corridors (to scale) and draw it in MATLAB using plot3. You may work in groups to get the dimensions, but list the group members in your documentation. We will extend this drawing in later assignments.

  3. Select an image to use as an input source for illustrating varying amounts of disortion from 10% to 90% in steps of 10%. See distort.m for a Matlab function that perform distortion.

  4. Repeat exercise 4 from the previous assignment (if you did not quite get it working) using code from the sample solutions as a starting point.

  5. Correct the three images of the front face of Kettering Lab (before rennovation) for distortion. The image KLWall1.jpg is shown below. Report the sign and magnitude of distortion used in each case. Do the same exercise for a photo from your web cam.

  6. Write a MATLAB script that uses ginput to get three arbitrary points and then produces the diagram shown below. As an extra challenge, superimpose this figure over an underlying image, using color to make the annotation stand out.

  7. Given three arbitrary 3D points, write a MATLAB script that calculates the unit normal vector to the plane determined by those points. The vector should be oriented by the right-hand rule. If your fingers curl from point 1 to point 2 to point 3, then your thumb should point in the direction of the normal vector.

  8. Generate a MATLAB plot (using plot3) of a 3D object and the orthonormal projection of that object onto a camera plane. As you rotate the 3D plot you should be able to see that the camera view is indeed a projection.

    beforeafter

  9. Find the autocorrelation images corresponding the textures in the folder patterns. Identify, where possible, the size and orientation of the unit cell. Repeat the exercise for several of your own images. Can an autocorrelation of an image ever be negative? Reference: auto.m. Note that this function assumes the input is a grayscale image. As a challenge you could modify it to work for color images.

Maintained by John Loomis, last updated 21 Feb 2011