Vector Quantization

(1)

Due Date: 12/14/2014

(2)

Download the Vector Quantization source code, vq.zip, from the class website

Use trvqsp_img to train a codebook for VQ

Use vqimg_enc and vqimg_dec to encode/decode images using the codebook you generated in the previous step

Use different parameters/images to test the performance of the VQ algorithm

2/4

(3)

There are three images in vq.zip: bookshelf, sinan, and sensin

In general, you should train your codebook using one image and encode a different image for

performance evaluation

As a starting point, try:

C:\> trvqsp_img sinan.raw codebook.dat -b 1024 -t 8 -w 8

The codebook.dat can be used to encode sensin as follows:

C:\> vqimg_enc –i sensin.raw –o data.bit –c codebook.dat

Finally, reconstruct the sensin image by

C:\> vqimg_dec –i data.bit –o sensin.raw

3/4

(4)

Please write a report including:

Description of the program flow of: trvqsp_img, vqimg_enc, and vqimg_dec

Try different parameters for VQ (e.g., codebook size, vector size)

Try different training images and compare the performance of VQ on different target images

Grading is based on

Your report

4/4