Page 164 - Algorithm Collections for Digital Signal Processing Applications using MATLAB

4. Selected Applications 153

Consider 23 texture images are stored in the database. They are indexed
using the low level features collected from the image itself, so that 23 texture
images are stored under 4 categories. K-means algorithm is used to classify
the collected texture into 4 categories so that each image in the database is
associated with the number indicating the category. (i.e.) 1 indicates first
category, 2 indicates second category and so on. This is called texture
grouping and is done as described below.

5.1 Approach

Step 1: Every texture image in the database is divided into sub blocks of
size 8x8. The variance of every sub blocks are calculated. They are
arranged in the vector form of size 1x64. This is called Low-level
feature vector extracted from the image itself. Note that elements of
the low level feature vector can also be any other statistical
measurements measured from the image.

Step 2: The collected Low-level feature vectors are subjected to k-means
algorithm to classify the images into four categories as described in
the chapter 2

Step 3: Thus the index number is identified for every image in the database.
Also the centroids of the individual category are also stored. The
centroid is the vector which is of the same size as that of the feature
vector. [Refer chapter 2]

Step 4: To retrieve the images from the indexed database that looks like the
image which is given as the key image is described below.

• Extract the low level feature vector from the key image as described in
the step 1.
• Compute the Euclidean distance between the computed feature vector
and all the centroids corresponding to the individual category. The
category corresponding to smallest Euclidean distance is selected.
• All the images in that category is retrieved using the index number
associated with every images