Contribution for the University and other Institutions
The research work performed during the last 14 years in the department in the field of Image Processing, Biometrics, Bio-Informatics, & Software Engineering found to be new and challenging. It has started with studying and analyzing the different aspects and prospective to solve the real world problems and gives appropriate findings.
The basic objective of the medical imaging research work was to provide the optimum solution in terms of quality medical images, for perfect diagnosis. The research work carried out by collecting the medical image databases from the local medical institutes and processed these images using the Image Processing techniques with optimum H/W & S/W cost in comparison to existing medical imaging system available in market.
While handling medical image enhancement short comings found in the process of early detection of the breast cancer, in screening mammogram can be reduced by newly developed image enhancement algorithm using heterogeneous approach (i.e. Morphology, FFT etc.) which found helpful in computerized analysis and detection of cancers missed by radiologists. Similar approach has been studied to analyze ultrasonic images. About 10 articles in International conferences/Journal have been published. This work may extend to generate the critical databases of the various types of medical images for various types of disease diagnosis.
In an increasingly digital technology world, among the main innovation prospects and framework of future communication systems, design of database access integral services, e-commerce, remote control of terminals and devices being the result of global services derived from last generation. Some features standout from such future services like authentication in human machine interacting to deal with security and identification problems, that’s why the use of biometric based technology get developed. This is new and emerging technology due to its high degree of maturity and reliability.
Fingerprint The performance of fingerprint recognition system is depends on the quality of input fingerprint image, so huge efforts are required to improving the quality of fingerprint image. In this work the composite algorithm, a novel SWT based Composite methods has been proposed for the Fingerprint Image enhancement. Also compared the performance of before and after of applying these algorithm by extracting features in terms of sensitivity and specificity. This result also reflects the average sensitivity that was 89% and 95% before and after post processing respectively. Similarly, for specificity, we got high specificity after pre-processing i.e. 89% as compare to before i.e. 78%.
Recently moved towards the Face Recognition and the hybrid biometrics (like Face & fingerprint, multiple fingerprints of different fingers etc)
Free style handwriting recognition is a difficult problem, not only because of the great amount of variants in handwriting, but also because of the overlapping and the interconnection of the neighboring character. In this work the development of an investigation of segmentation technique, structural based character classification. The box fitting method has been proposed and it gives 80% of recognition.
Computational simulation of experimental biology is another application for bioinformatics which is apply to refer to as ‘in silico’ testing. This perhaps an area, which will expand in prolific way, given a need to obtained a greater degree of predictability in animal and human clinical trials. Added to this, is interesting scope that in silico’ testing provides to deal with the brewing hostility towards animal testing.
The sequence based searching and explorations of the biological databases (Plant viruses) are often saves a lot of time in the lab. Identifying homologous sequences provides bases for phylogenetic analysis and sequence pattern recognition. Software for prediction for ORF, genes, exon splice sites, promoter binding sites, repeat sequences, and tRNA gene helps molecular biologist make sense out of unmapped DNAs. Inadequate sequences and data are available on plant viruses, which need further expansion and revision. These sequence need to be characterized for their use in agribiotech and pharmaceuticals. Thus, data Intensive techniques such as high throughput screening and gene expression experiments demand methods to correlate large and diverse datasets. Genomic research requires analysis of large sequences and is heavily dependent upon bioinformatics computation where strict standards, requiring careful data selection for protein model building, followed by adequate testing and validation through bioinformatics tools required.
Bioinformatics research is very large database with complex queries and essential requirements for reliability, availability, and scalability of sequence analysis. Under this study around 1837 Plant virus databases has been studied as per the above mentioned aspect and published 7 articles in world class journals. The work may continue further for gene prediction and drug designing under this scenario.
In this work we had studied the various aspects of traditional and Object Oriented Software Engineering in view of S/W Reliability, various S/W Reliability model has been studied in view of trend analysis, Defect Density, exponential and Logarithmic Model Parameters. It is found that the existing S/W Reliability Models are used late in the implementation phase or operational phase which seems too late to be utilized for enhancing the reliability of the S/W. Since changes made to the S/W design after the product is almost complete requires around 20% to 80% of additional cost. With this motivation we had studied the existing OO Quality models, frameworks and metrics in the early design phase and predicted the reliability focus quality model for OO design through various case studies. The practical exploration to this model using various test cases shows that the R-square and adjusted R-square close to one & zero value of P shows the model validity. The reliability estimation shows that as the CMM level goes on increasing the reliability go on increasing also the team level increases the reliability increases. About 5 articles in International conferences/Journal have been published.
Sponsored Major Research Projects
This project provides a comprehensive evaluation of fingerprint compression techniques using wavelet and wavelet packet, fingerprint image enhancement techniques, fingerprint image enhancement techniques using SWT, Fourier and spatial domain techniques and finally the feature extraction using mathematical morphology in terms of specificity and sensitivity to increase the robustness of the Automatic Fingerprint Recognition System (AFRS).
The main has been given to how compression ratio can increases by selecting appropriate threshold value. The compression ratio of noisy and noiseless fingerprint images are determining by considering number of zeros (NZ) and retain energy (RE) occurs after compression.
In this work in first phase we have applied Daubechies, Symlet and Coiflet transforms with different orders to fingerprint images of FVC2002 database. We determined NZ and RE at different orders from 1st to 5th level. If NZs are more, then compression ratio is also more. We found that to get highest compression ratio, the order and level must be high. We have compared these three transforms and their order as well as levels, and found that Coif let is most suitable for fingerprint image compression, because it gives more NZs at 4th order with 5th level. Hence for fingerprint image compression Coif let transform is most suitable.
In Second phase a novel SWT based Composite method has been used to deal with fingerprint image enhancement. The new approach adopts the de-noising by SWT.
The SWT based composite method provided better results on low contrast and noisy fingerprint images
Finally, we have introduced a method for removing superfluous information for genuine fingerprint feature extraction using mathematical morphological operation. This algorithm removes the spikes, spurs and dots very effectively and extracts a clear and reliable ridge map structure from input fingerprint image. We have also compared the performance of before and after of this morphological algorithm by extracting features in terms of sensitivity and specificity. This result also reflects the average sensitivity that was 89% and 95% before and after post processing respectively. Similarly, for specificity, we got high specificity after preprocessing i.e. 89% as compare to before i.e. 78%. About 28 articles in International conferences/Journal have been published.
Automatic pattern recognition system is essential to makes a personal identification by determining the authenticity of specific psychological characteristic (fingerprint) posses by the user to improve the system performance by considering performance evolution. At present so many work has done in biometric (Unimodal) recognition system, but these system have limitations in terms of noise in sensed data, intra-class variations, distinctiveness, non-universality (Failure to accept), spoof attack etc. cause poor performance. To overcome these problems the use of multimodal is substitute and influential solution. In this proposed work our object is to develop intelligent biometric techniques by taking multiple impressions (snapshot) and multiple matching algorithms (pattern recognition techniques i.e. advanced matching techniques).
Fingerprint matching is basically divides in two categories i.e. Minutiae pattern matching and Structural matching. Recently, observed that taking minutiae types into consideration while matching can increase system's accuracy. Therefore, minutiae types should be considered when high-quality images are being used and minutiae extraction algorithm is reliable. As per our study and reported work, it is observed that, due to high distortion of fingerprint image the biometric automatic recognition does not show higher performance of recognition even though in the well matured recognition system. So, probable solution for this problem is to take multiple fingerprint images as an input of an individual (multiple fingerprints of different fingers) or can also possible for heterogeneous matching algorithms. For fingerprints point pattern matching, graph based matching, fuzzy graph model generally used. Further we can extend this study by using advanced pattern recognition techniques, Neural Network, and fuzzy logic.