Digital Signal Processing Reference
In-Depth Information
algorithm. Mori et al. [10] segmented images into grids, clustered them, and obtained
the word distributions for a cluster from the keywords in it. Finally, they annotated an
input image by calculating suitable keywords from the distributions of the clusters
similar to the grids of the input image.
3
Proposed Algorithm for Annotated Images
The primary goal of the section is to select series keywords to describe un-annotated
image based on region features relevancy. First, using the image segmentation
algorithm produces the regions of each image, and then establishing the relevancy
between regions with keywords as a priori knowledge. Finally, through calculation
the similarity of un-annotated image regions with each training image's to annotate
the un-annotated image.
3.1
Image Segmentation Using K-Means Clustering Algorithm
Images can be segmented into regions by the K-means algorithm which is an
unsupervised learning algorithm that classifies a given data set into multiple classes
through a certain number of clusters based on their inherent distance from each other.
The algorithm assumes that the data features form a vector space and tries to find
natural clustering in them. The points are clustered around centroids
k
i
∀
i
=
1
k
which are obtained by minimizing the objective.
k
=
2
V
=
(
||
x
−
k
||
)
(1)
c
i
i
1
x
∈
K
c
i
K
, i=1,2,...,k and
k
is the centroid of all the points
Where there are k clusters
x
∈
K
As a part of this project, an iterative version of the algorithm was implemented.
The algorithm takes a 2 dimensional image as input. Various steps in the algorithm
are as follows:
a)
Compute the histogram of the intensities.
b)
Place K points into the space represented by the objects that are being clustered.
These points represent initial group centroids.
c)
Repeat the steps d-e until the cluster labels of the image does not change anymore.
d)
Cluster the points based on distance of their intensities from the centroid intensities.
c
i
(2)
c
(
i
)
=
arg
min
||
x
(i)
−
k
||
2
j
j
e)
Recalculate the new centroid for each of the clusters.
