Page 152 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

4. Selected Applications 141

2. EAR IMAGE DATA COMPRESSION USING
EIGEN BASIS

Collected ear images as mentioned in the section 1 are represented as the
group of 85x60 sized matrices. The elements of the matrices are stored with
the numbers ranging from 0 to 255.8 bits are required to store every number.
(i.e.) 8*85*60=40800 bits are required to store the single gray image.
Therefore there is the need to identify the technique for storing the image
data with reduced number of bits. This technique of representing the data
with reduced number of bits by removing the redundancy from the data is
called Image data compression.
Ear image data compression using eigen basis is the compression
technique which exploits psycho visual property of the eye. This technique is
used to compress the similar images belong to the particular group. In the
experiment described below, the group considered is the ear images
collected from many persons. The steps involved in Image data compression
using Eigen basis are described below.

2.1 Approach

Step 1: 8x8 sized subblocks of the ear image matrix are collected randomly
from the ear image database.

Step 2: Reshape the subblock matrix into the vector of size 1x64.

Step 3: Thus set of 100 vectors are collected randomly as described in step1
and step2.

Step 4: Co-variance matrix is computed for the collected vectors.

Step 5: Eigen values are computed for the covariance matrix. Eigen vectors
corresponding to the significant Eigen values are computed. They are
called Eigen basis, which spans the Ear vector space, which consists of
all the vectors available as the reshaped sub blocks in the ear images.
Note that Eigen vectors thus obtained are orthonormal to each other.

Step 6: Number of Eigen basis obtained as described in the step 4 is less
than 64 (vector size). This number is called as the dimension of the
Ear vector space.

Step 7: To compress the ear image, ear image matrix is divided into
subblocks of size 8x8. Reshape every sub block into the vector of