PROJECTS


I have, in the course of my Bachelor's degree of Computer Engineering, 
done the following projects :


 To other links


DATA  MINING



 Data Mining is a technology that  helps you identify and extract high-value business
 intelligence from data assets. Analysts and business technologists are empowered to 
 discover patterns which might otherwise be unobserved, across volumes of data they 
 were simply not able to penetrate with other types of analytical tools. Data Mining 
 provides the fundamental technology and tools to support the mining process, as well
 as application services to enable development of customised applications.


Innovative technology

 Central to this technology are its data mining algorithms.  Applicable to a wide range
 of business problems, the algorithms can be employed to facilitate decision-making in 
 business areas such as campaign planning, customer relationship management,
 process reengineering, product planning and fulfilment, and fraud and abuse management.
 
 Capabilities include support for the following methods:

    

  • Creation of classification and prediction models

  • Discovery of associations and sequential patterns in large databases 

  • Automatic segmentation of databases into groups of related records 

  • Discovery of similar patterns of behaviour within special time sequences 

  • Data analysis and preparation



 Automation of some of the most typical data preparation tasks is aimed at improving 
 the analyst's productivity by eliminating the need for programming  
 specialised  routines.  Depending on the data mining technique, the analyst can select,
 sample, aggregate, filter,   cleanse and transform data in preparation for mining.
 
 Statistical functions facilitate analysis and preparation of data as well as provide 
 forecasting  capabilities. Statistical functions supported include factor analysis, 
 linear   regression,  principal component analysis, univariate curve fitting, 
 logistic  regression,   univariate and  bivariate statistics.
 

Visualisation

 Visualisation functions bring out unusual features that might otherwise be missed. 
 Customised    visualisation interpret data mining results, depict model quality, and 
 present the  results of   various statistical functions.
        
Back to the top





Image Compression Using Vector Quantization




Introduction

    
 Digital storage of photo ID pictures ,along with other personal ID information is 
 required in various personal identification and security systems.Such systems can 
 either be centralized ( the image information is stored in a central database) or 
 decentralized (the image of each individual is stored in his own ID card) In the 
 latter case the memory is very limited. One example of the storage media on an ID 
 card is  the two dimensional barcode known as PDF417 label, which has the capacity 
 of 1024 bytes. An 8-bit monochrome facial image at a spatial resolution of 128x128 
 pixels occupies 16,384 bytes. A compression factor of more than 90 % is required for 
 such an image to be compressed down. At high compression ratios and when starting from
 low resolution,Vector Quantization (VQ) algorithm can yield a good image with an average
 compression of more than 90 %. 
 
 Image data  compression techniques fall into two broad categories.In the first category,
 called predictive coding, are methods that exploit redundancy in data. Redundancy is a 
 characteristic related to the factors as predictability, randomness and smoothness in the 
 data e.g., an image of constant gray levels is fully predictable once the gray level of 
 the first pixel is known. On the other hand, a white noise random field is totally 
 unpredictable and every pixel has to be stored to reproduce the image. Techniques such 
 as Delta Modulation and Differential Pulse Code Modulation fall in this category.

 In the second category, called transform coding, compression is achieved by transforming 
 the given image into another array such that a large amount of information is packed into
 a small number of samples. Other image data compression algorithms exist that are
 generalisations or combinations of these two methods. 

 This project deals with storage and compression of images ( photo IDs) using 
 Vector Quantization .



Vector Quantization

 Vector Quantisation (VQ) can be defined as a form of pattern recognition where an input
 pattern is "approximated" by one of a set of patterns.It is mostly applied to image 
 compression. An input image is first divided into meshes. Each mesh is then approximated
 by one of a set of patterns. We can call a mesh an input by one of  a set of patterns. We
 call  a mesh an input Vector and a set of patterns a code book. A vector in a codebook is
 reffered to as a code vector. For VQ to be put into use two issues need to be addressed.
 One is the accelaration of the " nearest neighbor search (NNS)". The nns looks for a vector
 nearest to an input vector among a large number of input vectors. This requires a substantial
 amount of time on convential serial processors in relation to the dimension and the number 
 of code vectors. The other issue is the deseign of an optimal codebook. The quality of
 reconstructed images obtained from a common set of images using DCT-based compression 
 algorithms is independent of the algorithm used. In VQ, however, the quality of the 
 compressed image depends on the codebook design.



Results


The results are as shown below:






This screen gives the options to the user




This is the screen when user selects option 3 above





This is the compressed image (90 % compression) using 2x2 blocks and 128 entries

in the codebook

94.1 % has been achieved with 4x4 blocks and 256 entries of codebook






This gives the mapping of 256 colors to 16 colors for another quantization scheme implemented






The result of above quantization scheme





Back to the top





Home   About Me   College    Resume    India     U of A    Goals   Contact Me