You are on page 1of 4

Course: CSCE6370

Name: VenkatSaiSailesh Chennam

Assignment: RA#01

File name: RA01_Chennam.doc

Date: 02/08/2011

Indexing and Retrieval Medical Images Based on 2x2 DCT and


IDS Compression
Introduction:
This paper is all about Content Based Medical Images Indexing and Retrieval (CBIR) techniques
using 2x2 Discrete Cosine Transform (DCT) and Information Dominance Strategy (IDS). The
paper conceptualizes about compressed image database coupled with Content Based Images
Indexing and Retrieval (CBIR) techniques that are mainly used for retrieval of similar images of
the query image implemented by the query tools.

Existing System (In past years):


Previously image indexing and image database assembling used to be text based that is typing a
similar key word for retrieval of similar images but eventually it became evident that many
things are hard to express such as feelings, situations especially what is in the query image and
what is it all about.

Therefore for the existing system it was clear that there was a need for combination of textual as
well as visual indexing techniques which is now referred to as Content Based Indexing and
Retrieval (CBIR) system.

Present System:
Content Based Images Indexing and Retrieval (CBIR) techniques which are based on features (or
attributes) such as histogram, color, shape etc.
Author’s Proposal on DCT:

J R Smith:
According to J R Smith an image is divided into 4x4 blocks and DCT is computed for each block
resulting in 16 coefficients. The feature vectors are generated by calculating variance and mean
absolute values of each of these coefficients.

R Reeves:
Unlike J R Smith who divided the image into 4x4 blocks, R Reeves proposed a DCT based
texture discrimination technique where in an image is divided into 8x8 blocks and the feature
vector is formed with variance of only first 8 AC coefficient thereby decreasing the time
complexity when compared to J R smith’s time complexity.

M Shneie:
M Shneie proposed a technique for image retrieval using JPEG. This is based on mutual
relationship between DCT coefficients of unconnected regions of both query image and the
target image.

Currently multiresolution techniques such as DWT and subband analysis are proposed for
Texture Based Image Retrieval. According to researches, algorithms using multiresolution
techniques can achieve very good performance on texture analysis and obtain much better
correct classification rate and retrieval accuracy than other typical image decomposition such as
DCT and spatial partitioning.

Implementation of 2x2 DCT and IDS process for Image Indexing and
Retrieval:

Introduction:
As we have seen earlier that a JPEG image is divided 4x4 or 8x8 DCT coding size. This becomes
a disadvantage when large blocks of transformation encoding are done for a full frame DCT.
Therefore a 2x2 DCT is used which has the ability to retain all the original characteristics of such
large image ( like JPEG) thereby reducing the computational complexity and memory
requirements.

Procedure:
Initially the image is sub divided into 2x2 blocks and the translation function T is applied. This
decorrelates the spectrum data and each spectral data is collected individually, resulting in four
different frequency domains which when clustered becomes a multiresolution image. Here the
block with lowest band indicates the average of 4x4 grey data and the other bands contain
contour information of original image data. A second 2x2 DCT and function T are applied to the
lowest band thereby further decreasing the spectrum complexity.

IDS Process:
The IDS process is employed for two main reasons

(i) Select the dominant(more information) block


(ii) Decrease the transmission overhead of position information of dominant sub blocks
selected.

Extraction of Feature Vectors:


Yu-Hen proposed a method for feature extraction based on rearrangements of DCT coefficients.
The four coefficients within the block are stored into four information subbands. Energies are
measured by evaluating the Variance Vk and Mean Absolute Value Mk.

Similarity between query image (IR) and the target image (IT) is given by cosine distance

Performance Quotient:

The performance of the retrieval system can be evaluated by the retrieval efficiency .

Where nq is number of successfully


retrieved images.

Conclusion:
In this paper we have studied two types of image indexing and retrieval techniques. Many
authors have proposed image indexing and retrieval in compressed domain which uses low level
features such as color, texture, shape etc but image indexing and retrieval which is texture based
using multiresolution techniques such as subband analysis will intensify the performance
quotient by giving much better classification rate and retrieval accuracy DCT .
References:
[1] http://cobweb.ecn.purdue.edu/RVL/Research/CBIR/index.html

[2] http://www.clef-campaign.org/2005/working_notes/workingnotes2005/lim05.pdf

[3] http://thomas.deselaers.de/teaching/files/tutorial_icpr08/03_textBasedRetrieval.pdf

[4] http://www.springerlink.com/content/h023071202j1404j/

[5] http://www.ee.columbia.edu/~sfchang/course/svia/papers/rui99_cbir_survey.pdf

You might also like