MultimediA

You need to write the code yourself in Matlab

Use of publically available code (e.g., from online sources) is not permitted

Plagiarism will be severely penalized!

Multi view image encoder

Multi-view image decoder

Project deliverables? Documented MATLAB software

• Comments within .m files explaining the code
• Project report
• Microsoft Word (+ Equation Editor)
• Introduction
• Background
• Details of your project
• Results
• References (IEEE format)
• Explanation of your Matlab code (User manual)

Presentation

Towards the end of the course, some people may be asked to present their codec design

If asked to present, please prepare a 10-15 minute talk

PowerPoint slides

Codec demo

Matlab software?

• Your software should have at least two functions:
• im_encode(name, bitstream_name, N_images, parameter(s)) im_decode(bitstream_name, dec_name, N_images)
• where
•  name is the common prefix of the image filenames (e.g., for ‘aloe0.pgm', ‘aloe1.pgm',..., name = ‘aloe')
•  N_images is the number of images
• parameter(s) stands for any parameters that you wish to use in encoding (e.g., quality, ...)
•  dec_name is the common prefix of the filenames of decoded images (e.g., ‘aloe_dec' for the example above)
•  bitstream_name is the compressed bitstream filename
•     SFU ENSC 424 8
•  Of course, your software may have other functions as well, depending on the transform, quantizer, and entropy coder you wish to use, but im_encode and im_decode are essential
•  Important: your encoder should be able to produce different image qualities and different bit rates by changing the value(s) of parameter(s) (e.g., quality)
•  Important: any information passed from the encoder to the decoder must be in the compressed bitstream (no information passing through Matlab workspace). In other words, you should be able to encode the stereo image pair using im_encode, then exit Matlab, open Matlab again, and then decode the bitstream using im_decode.

SFU ENSC 424 9

• Several multi-view image sets will be provided on WebCT for practice
• Each set contains up to 10 images with the following characteristics:
• Grayscale (i.e., not color)
• Filenames in the form: ‘.pgm', for example ‘aloe0.pgm', ‘aloe1.pgm', .etc.
• Sizes of images in the same set will be the same, but will vary between image sets; for example, ‘aloe' images are of size 424×368, while ‘art' images are of size 464×368
•  Dimensions of all images will be multiples of 8 pixels

SFU ENSC 424 10

• The performance of your codec will be evaluated based on rate-distortion criteria
• Bit rate = (total # of bits in the compressed file(s)) divided by the (total number of pixels in the N images)
• Distortion is the Mean Squared Error (MSE)

Benchmark performance - aloe:

n      The higher the curve – the better

n      Project marks determined, to a large extent, by competition among codecs

Benchmark performance - art