You are on page 1of 1

FORMULATION OF PROJECT AND PLAN FOR IMPLEMENTATION:

We have planned to use a image coding technique called vector quantization to extract the feature vector
of an image as it requires less computational power. Extracted feature vector is compared with the feature
vector of the images stored in the database and the recognition rate is calculated. Getting more
improvement in the recognition rate is the major aim of the project.
Implementation consists of following steps:
1) Collect the image database.
2) Apply low pass filtering to reduce high frequency noise.
3) Divide the filtered image into 4x4 pixel blocks.
4) Find minimum intensity in individual 4x4 block and subtract it from each pixel in the block.
5) Apply vector Quantization to each block to obtain feature vector. 
6) Generate codebook and compare it with registered codebook to find nearest match.
7) Count matched frequencies for each codevector and generate a histogram.
8) Modify the above procedure to improve recognition rate.
For step 5 and 6 apply following steps:
Traineimage folder contains 400 images of 50 persons.(8 images of each person.)
They are numbered as tr1 to tr8 for person1, tr9 to tr16 for person2 and so on. 
Testimage folder contains 350 images of 50 persons.(7 images of each person.)
They are numbered as te1 to te7 for person1, te8 to te14 for person2 and so on. 
We have to test each image in test folder against each image of trainee folder.
Use LBG algo for vector quantization.
Find feature vector of each color image by considering three separate planes R,G,B and combination of
these three feature vectors gives the complete feature vector of one image. Arrange them in matrix form
for comparison.
First use manhattan distance as a measure of similarity and then eucledian distance as a measure of
similarity.

You might also like