10-03-2011, 12:47 PM
Presented by:
E.Malleswara Rao
[attachment=9941]
Image Segmentation Using Information Bottleneck Method
The purpose of this paper is to introduce new segmentation algorithms using a hard version of the information bottleneck method. The objective of this method is to extract a compact representation of a random variable with minimal loss of mutual information with respect to another variable.
Problem Definition:
Existing system:
New segmentation algorithms using a hard version of the information bottleneck method is considered. The use of this method requires the definition of an information channel where a random variable controls the clustering of the other by preserving the maximum mutual information between them.
Proposed Work Overview:
Three different segmentation algorithms have been introduced: a split-and-merge, a histogram clustering and a registration- based clustering. For the two first algorithms, an information channel between the regions of the image and the histogram bins has been defined. For the third algorithm, a channel between two multimodal images is defined.
Proposed work:
The following information-bottleneck-based algorithms represent the main contributions of this paper.
1) Split-and-merge algorithm: In the first phase, a top-down strategy is applied to partition an image into quasi-homogeneous regions using a binary space partition (BSP) or a quad tree partition. In the second phase, a bottom-up strategy is used to merge the regions whose histograms are more similar.
2) Histogram clustering algorithm: Neighbor bins of the histogram are clustered from a previously partitioned image. After assuming that the split-and-merge algorithm provides us with the structure of the image, our clustering algorithm tries to preserve the correlation between the clustered bins and the structure of the image.
3) Histogram clustering algorithms for two registered multi- modal images: Two different algorithms are presented. The first one segments just one image at a time, while the second one segments both simultaneously. The clustering process works by extracting from each image the structures that are more relevant to the other one. In these algorithms, each image is used to control the quality of the segmentation of the other.
Input: where the input variable represents the histogram bins
Output: The output is given by the set of regions obtained from the above split-and-merge algorithm.
Implementation:
Software Requirements:
The major software requirements of the project are as follows.
Language : Mat lab 7.0
Operating System : Windows XP/Vista
Hardware Requirements:
The Hardware requirements that map towards the software are as follows
Processor : Core2Duo/Dual Core
RAM : 1 GB
Hard Disk : 80 GB