198:443 Digital Imaging and Multimedia

Spring 2008

Assignment 2

Due Date – in class: Thursday March 6th

Submission instruction:

1- Report: You need to prepare a report containing all your answers, results, images, plots, and codes in one document. For example, you can use MS Word or any similar software to create a document and copy and paste the resulting images into it. Print the report and submit a hard copy of it as well as email it to the TA.

2- Code: You also need to submit an electronic copy of your code.

Zip all your code in one file and email it to the TA.

3- You can use ImageJ or Matlab for your implementation.

- Late submission policy: 20% off the grade for each day late.

The images for this assignment are available at

ftp://ftp.cs.rutgers.edu/pub/elgammal/CS443/HW2

[Part I: Theoretical questions - 10pts each]

Solve problems Pr 6.1, 6.2, 6.6, 6.9, 7.1

[Part II:]

[Q II-1 – 40pts] Read section 5.8

Point operations on two images can be used to detect moving objects by subtracting consecutive images in an image sequence. In this problem we will use a sequence of operations to detect soccer player in a field from two frames. Use the images soccer1, soccer2 for this problem. In general we call them images I₁, I₂. Use ImageJ to implement the following steps:

1- Given two images I₁, I₂, use ImageJ point operation to obtain the absolute difference image d=abs(I₁-I₂).

2- Plot the histogram for the absolute difference image d.

3- Visualize the absolute difference image d. To get a good visualization, you need to stretch the range of values in d to be from 0-255. Use a point operation to achieve this stretching. Show the result.

4- Threshold the difference image d by using a suitable threshold a_th [Eq 5.5] to obtain a binary image (mask) where the moving pixels is set to a value 1 and the stationary pixels is set to 0. Choose a threshold such that p% of the pixels are above the threshold. Try with p=2%, 1%, 0.1%, 0.01%. In each case show the resulting mask. The mask should show the moving objects in the scene. Which p gives you the best results qualitatively?

5- We can obtain better results by using morphological close operation. Use ImageJ close operation with suitable window size. You can try two different options here

I. Use binary morphological ‘close’ on the mask m.

II. Alternatively, use gray-level morphological ‘close’ on the difference image d before thresholding to obtain a better mask

Which of the above methods give you better results? You don’t need to show results with different thresholds here. Just show the results with suitable threshold.

6- Use the obtained mask to visualize the moving pixels. Let m be the binary mask obtained in step 4. We need to obtain image I_m such that:

I_m(u,v)= I₂(u,v) if m(u,v) =1

0 if m(u,v) =0;

Show the image I_m showing the best result you get using the different options and thresholds from part 4 and 5.

- The same approach can be used in surveillance to detect moving people and vehicles. Re-run your code on images lot1, lot2 to detect the people in the parking lot scenes. Show the final result.

[Q II-2 - 25pts]

Point operations on multiple images can be used to add the effect of motion blur given a sequence of images. In this problem we will use alpha blending (page 83) on an image sequence to obtain a motion blur effect. Let I₁,I₂, …., I_T be a sequence of images. We can recursively blend the images into one image B as follows

Initialize B:

B= I₁

Recursive blending:

B = alpha B + (1-alpha) I_t , t=1,…,T

Use ImageJ to implement the above blending of multiple images. Use the image sequence 100.tif,…, 110.tif to test your code. Try with different alpha values= 0.1, 0.2, 0.5, 0.8. Show the resulting image in each case.

[Q3 – 15pts] Solve Exercise 6.10. Use ImageJ to implement the horizontal blur filter and test your code on the image 100.tif