Page 182 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

4. Selected Applications 171

8.1 Approach

• The 10 photographic images and the 10 photo realistic images are
collected (see figure 4-3 and figure 4-4). The sub blocks of the image
sized 16x16 are randomly collected from both the categories. 500 sub
blocks are collected from photographic images and 500 subblocks are
collected from the photorealistic images.

• The subblock thus obtained is reshaped into the size 1x256. Auto
Regressive (AR) co-efficients of size 1x10 are obtained from the
reshaped subblock. This is repeated for all the collected sub blocks. The
set of AR vectors collected from photographic images and the photo
realistic images are subjected to ICA analysis and 10 ICA basis each of
size 1x10 are obtained. [Refer chapter 2]

• Every AR co-efficient vector obtained from the particular subblock of the
images is represented as the linear combination of 10 corresponding ICA
Basis. The co-efficient of the ICA basis are obtained using the inner
product of ICA basis with the corresponding AR co-efficients. This is
called feature e vector of that particular subblock of the image.

• Thus 500 feature vectors are collected from photographic images and the
500 feature vectors are collected from the photorealistic images. The
centroid of the feature vectors collected from the photographic images is
computed as C1. Similarly the centroid of the feature vectors collected
from the photo realistic images are computed as C2.

8.1.1 To classify the new image into one among the photographic
or photorealistic image

The image is divided into sub blocks. Feature vectors are extracted form
every sub blocks using ICA basis as described above. The Euclidean
distance between the feature vector obtained from the particular subblock
and the centroids ‘C1’ and ‘C2’ are computed as ‘d1’ and ‘d2’ respectively.
Assign the number 1 to that particular subblock if ‘d1’ is lowest Otherwise 2
is assigned to that subblock. This is repeated for all the sub blocks of the
image to be classified.
Count the number of ‘1’s and the number of ‘0’s assigned to the
subblocks of the image to be classified. If the number of 1’s is greater than