Vandana Dixit Kaushik

An Efficient 3D Face Recognition Algorithm

This paper proposes an efficient expression invariant algorithm for 3D face recognition of subjects. The proposed algorithm uses a surface-based approach to extract automatically and to define geometrically the facial features like eyebrows, nose and lips. These extracted features are used to obtain some unique control points that can be used for matching. The algorithm is tested on Binghamton University 3D Facial Expression (BU-3DFE) database where each subject is having at least 6 expressions and is found to be more than 98.5% accurate.

An efficient indexing scheme for face database using modified geometric hashing

Abstract This paper presents an efficient scheme to index the database of facial images. It has made use of the modified geometric hashing technique. It uses minimum amount of search space and memory to provide top best matches with high accuracy against a query image. Control points are extracted using Speeded-Up Robust Feature points (SURF) operator. A pre-processing technique consisting of mean centring, principal components, rotation and normalization has been proposed to make these control points invariant to translation, rotation and scaling. The modified geometric hashing is used to hash these control points to index of the hash table. The indexing scheme has been tested on FERET face database which has achieved 100% hit rate for top 4 best matches.

An efficient natural image deblurring algorithm

This paper has considered the problem of deblurring of an image which is an ill-posed and challenging problem due to not only the large number of unknowns but also non-availability of more number of images of the same scene or objects. It uses the Variational Bayesian approach to optimize the posterior probability and to derive the most probable Point Spread Function (PSF). Lucy Richardson algorithm [1] has been modified to get the deblurred image. The algorithm is found to be very effective for natural images.

Comparative Analysis of Contrast Enhancement Techniques of Different Image

Contrast enhancement is an important technique of image processing for the enhancement of contrast of an image in the spatial domain. Numbers of contrast enhancement techniques are available. This paper proposes comparative analysis of various contrast enhancement techniques applied on different images. PSNR and MSE value after enhancement is depicted in the tabular form as well as graphical form to analyze the best enhancement technique applied on corresponding images.

Geometric Modeling of 3D-Face Features and Its Applications

This paper proposes a method of geometric modeling of features extracted from 3D face and presents some of its applications. It describes a new automatic pose and expression invariant feature extraction algorithms to extract features (control points) from eyebrows, nose and lips of 3D facial data. The proposed algorithms are tested on BU-3DFE (Binghamton University 3D Facial Expression) Database where each subject has at least six expressions. Three dimensional curves are fitted on these extracted features. Through Chi-Square test it reveals that the curve fitted against each extracted feature is found to be good with 98.5% confidence level. In this paper, the proposed geometric model has been used in 3D face recognition and in regeneration of all features. It has been found that the model helps to reduce the storage space considerably and can be used as a soft biometric tool to classify the biometric images in the database so that search space in the database for identification can be reduced substantially.

Automatic 3D Facial Feature Extraction Algorithm

This paper presents a novel algorithm for automatic facial feature extraction from 3D facial data. The proposed algorithm is found to be pose and expression invariant. It is successfully tested on 3D face database from the State University of New York, Binghamton, USA. In order to test the algorithm, the images have been regenerated using these extracted features. Bezier surface algorithm has been used for regeneration.

An Adaptive Multi-algorithm Ensemble for Fingerprint Matching

Any well known fingerprint matching algorithm cannot provide 100% accuracy for all databases. One should explore the possibility of fusion of multi-algorithms to achieve better performance on such databases. One of the major challenges is to design a fusion strategy which is both adaptive and improving with respect to the candidate database. This paper proposes an adaptive ensemble using statistical properties of two well known state-of-the-art minutiae based fingerprint matching algorithms to achieve (1) improvement on fingerprint recognition benchmark, (2) outperform on multiple databases. Experiments have been conducted on two databases containing multiple fingerprint impressions of 140 and 500 users. One of them is widely used publicly available databases and another one is our in-house database. Experimental results have shown the significant gain in performance.

A Hybrid Image Compression Technique for Medical Images

Due to the bandwidth and storage limitations, medical images must be compressed before transmission and storage. However, the compression will reduce the image fidelity, especially when the images are compressed at lower bit rates. The reconstructed images suffer from blocking artifacts and the image quality will be severely degraded under the circumstance of high compression ratios. In this paper, we present a strategy to increase the compression ratio with simple computational burden and excellent decoded quality. Higher compression ratio is achieved by applying different compression thresholds for the wavelet coefficients of each DWT band (LL and HH) while DCT transform is applied on (HL and LH) bands with preserving the quality of reconstructed medical image. The retained coefficients are quantized by using adaptive quantization according to the type of transformation. Finally the entropy coding (variable shift coding) is used to encode the quantization indices. The Discrete Wavelet Transform (DWT) analyzes the signal at different frequency bands with different resolutions by decomposing the signal into an approximation and detail information. Image coded by DWT do not have the problem of blocking artifacts which the DCT approach may suffer.

An Efficient Fingerprint Indexing Scheme

This paper proposes an efficient geometric-based indexing scheme for fingerprints. Unlike other geometric-based indexing schemes, the proposed indexing scheme reduces both memory and computational costs. It has been tested on IITK database containing 2,120 fingerprints of 530 subjects. Correct Recognition Rate is found to be 86.79 % at top 10 best matches. Experiments prove its superiority against well-known geometric-based indexing schemes.

Curvelet based image fusion

This paper presents an efficient image fusion algorithm which is based on curvelet transform. It makes use of different characteristics of the same geographical area obtained from more than one image, to obtain a more informative image. Experiments have been carried out to fuse monochromatic PAN (PANchromatic) satellite image with a multispectral LISS (Linear Imaging and Self Scanning) III image and it has been found that the fused image contains much richer information in both the spatial and spectral domain simultaneously.