EE 520 Topics in Digital Image and Video Communications (Spring 2000)

  • Course description

  • Syllabus

  • Reading list

  • Standard images for experiments

  • Software Programs

  • On-line Papers (IEEE etc.)

  • Projects

  • Homework 1

    • Compute the zero-order entropy of the images in the dataset. Compress the images using the Huffman coder and tabulate the compression ratios. Explain the differences in the compression ratios. Decode the Huffman encoded image to check the implementation.

  • Homework 2

    • (a) Show that the Gaussian random variable has the maximum differential entropy compared to any other continuously distriubuted random variable with the same variance.
    (b) Write a program to compute the histogram of a 8 bit/pixel grey level image of any size.
    (c) Encode an image from the dataset using a forward adaptive uniform quantizer. Choose block size=8 by 8. Compute the peak SNR (in dB) for coding rates (R) = 1,2,3,4 bits/pixel and plot the block number vs step-sizes. Compare the peak SNRs with the peak SNRs of non-adaptive uniform quantizers of same rates.

  • Homework 3

    • (a) Implement the pdf-optimized Lloyd-Max quantization algorithm. The implementation should be able to handle any pdf. Test the implementation by comparing your results with Max's paper for a four-level and seven-level quantizer when the pdf is zero mean, unit variance Gaussian. Compute the SNR (in dB) and entropy of the output of the quantizer. Generate 500 samples using a zero-mean, unit variance Gaussian random number generator. Quantize these samples using the four-level and seven-level quantizers that you computed. Plot the orignial samples over the quantized samples. You are encouraged to use the Matlab routines provided in class.
    (b) Implement the Laplacian pyramid based progressive transmission scheme based on the paper given in class. Reproduce Fig. 4, Fig. 5, Fig. 6, Fig. 7, Fig. 8, and Fig. 9 given in the paper. Use Lenna image.

  • Homework 4

    • (a) Find the first-order correlation coefficient of Lenna, Barbara and Man images. Compute the correlation coefficient matrix for Lenna (size 32 by 32).
    (b) Construct a VQ keeping the number of codewords fixed and choosing block sizes : 4x4, 8x8 and 16x16. Plot the Rate Vs Distortion (R-D) curves for these. Choose the codebook sizes to be 32, 64, 128, 256. Comment on the curves. (Code the Lenna image with these codebooks. Pick your choice of images for the training set. It must be distinct from the test image (Lenna)). Comment on your choice of training images.

  • Homework 5

    • Show that orthogonal transforms are variance preserving. That is, if \sigma_x^2 is the variance of the zero-mean input signal, X, and \sigma_k^2=E(\theta(k)^2) is the variance of the kth transform coefficient, k=0,1,...,N-1 then the average value of {\sigma_k^2; k=0,2,...,N-1} is equal to \sigma_x^2.

  • Homework 6

    • Use and Lenna and Barbara for the following. Compute 2-D DCT for the image using 16x16 block sizes.
    (a) Plot the R-D curve when, only 10%, 30%, 50%, 70% and 90% of the highest energy (equal to DCT co-efficient (magnitude)^2) DCT coeffs are retained in each DCT block, i.e. the other DCT co-effs are quantized to zero. Find the PSNR and rate if only the DC co-effs. are retained.
    (b) Quantize the DCT coeffs. in each 16x16 block using a 4, 8, 16 and 32-level uniform quantizer. Encode the output of the quantizer using a Huffman coder. Compare the compression ratios with and without Huffman coding for each quantizer.
    (c) Quantize the DCT coeffs (block size 16x16) using uniform quantizers. Compute the rate for these quantizers using the optimal bit allocation algorithm when the average rate is 1 bit/pixel. Find the PSNR and show the bit allocation matrix.

  • Homework 7

    • Download the "Walter Cronkite" and "Toy Vehicle" video frames from this website .
    (a) Write a program that will display the frames as a video movie (just like the one given in the above website). Use any programming language/tool of your choice.
    (b) For these two video sequences find the motion vectors using Frame 1 and Frame 2. Also, find the motion vectors using Frame 1 and the last frame in the corresponding video sequence. Use Frame 1 as the reference frame. Use the three-step search algorithm given in the class notes for motion estimation. Use block sizes of 8x8 and 64x64 and compare the results. You can use the "quiver" function in MATLAB to display the motion vectors on the image blocks. The grade for this homework will be based on the demonstration of your implementation. You are NOT required to submit a hard copy of your results.

  • Homework 8

    • In your opinion what are the properties a digital watermark should satisfy to resolve rightful ownership issues. Explain.

  • Homework 9

    • Use the spatial watermarking technique discussed in class to study its resilience against a collusion attack. What is the effect of the magnitude of the embedding factor on the collusion attack ?


     


    Fall 1999

    Main Page
    ECE