Computer Science - The City College of New York
CSC I6716 - Spring 2015  3D Computer Vision

Assignment 1 ( Deadline: March 03 before class)


Note: All the writings must be hard copies in print - you may hand them in during class, or put them in Prof. Zhigang Zhu's mailbox in the CS main office. . You also need to turn in your soft copies of your assignment by sending Prof. Zhu by email attachments. You are responsible for the lose of your submissions if you dont include  CSC I6716 (exactly) in the subject of your email. Send your source code ONLY please dont send in your images and executable. Do write your names and IDs (last four digits) in both your hard copy and soft copy submissions.

1. Writing Assignments (10x4 = 40 points)
(1). How does an image change (e.g., objects' sizes in the image, field of view, etc.) if the focal length of a pinhole camera is varied?
  (2). Give an intuitive explanation of the reason why a pinhole camera has an infinite depth of field.
  (3). In the thin lens model, 1/o + 1/i = 1/f, there are three variables, the focal length f, the object distance o and the image distance i (please refer to Slide # 18 of the Image Formation lecture). If we define Z = o-f, and z = i-f, please write two a few words to describe the physical meanings of Z and z, and prove that Z*z = f*f given 1/o + 1/i = 1/f.
  (4). Prove that, in the pinhole camera model, three collinear points (i.e., they lie on a line)  in 3D space are imaged into three collinear points on the image plane. You may either use geometric reasoning (with line drawings) or algebra proof (using equations).

2. Programming  Assignments (Matlab preferred - here is a quick matlab tutorial.  You may use C++ or Java if you like, but you will need to bring your  own machine to me in my office hours to run your programs. ) (10x6 = 60 points)

Image formation.  In this small project, you are going to use Matlab to read, manipulate and write image data. The purpose of the project is to make you familiar with the basic digital image formations. Your program should do the following things:

  1. Read in a color image C1(x,y) = (R(x,y), G(x,y), B(x,y)) in Windows BMP format, and display it.
  2. Display the images of the three color components, R(x,y), G(x,y) and B(x,y), separately. You should display three black-white-like images.
  3. Generate an intensity image I(x,y) and display it. You should use the equation I = 0.299R + 0.587G + 0.114B (the NTSC standard for luminance).
  4. The original intensity image should have 256 gray levels.  Please uniformly quantize this image into K levels ( with K=4, 16, 32, 64).  As an example,  when K=2 ,  pixels whose values are below 128 are turned to 0,  otherwise to 255.  Display the four quantized images with four different K levels  and tell us  how the images still look like the original ones.
  5. Quantize  the original three-band color image C1(x,y) into K level color images CK(x,y)= (R(x,y), G(x,y), B(x,y)) (with uniform intervals) , and display them. You may choose K=2 and 4 (for each band).  Do they have any advantages in viewing and/or in computer processing (e.g. segmentation)?
  6. Quantize  the original three-band color image C1(x,y) into a color image CL(x,y)= (R(x,y), G(x,y), B(x,y)) (with a logarithmic function) , and display it. You may choose  a function  I' =C ln (I+1) ( for each band), where I is the original value (0~255) , I' is the quantized value,  and C is a constant to scale I'  into (0~255), and ln is the natural logarithm.  Please find the best C value so for an input in the range of 0-255, the output range is still 0 - 255. Note that when I = 0, I' = 0 too.
Please for each of the above, provide your observations / conclusions, rather than just show the experimental results.
I have provided a piece of starting code for you to use. Questions a and b have been done.  You only need to work on c to g. You may use Prof. Zhu's old ID picture for testing your algorithm.