Ilaiah Kavati

Vein Pattern Indexing Using Texture and Hierarchical Decomposition of Delaunay Triangulation

In biometric identification systems, the identity corresponding to the query image is determined by comparing it against all images in the database. This exhaustive matching process increases the response time and the number of false positives of the system; therefore, an effective mechanism is essential to select a small collection of candidates to which the actual matching process is applied. This paper presents an efficient indexing algorithm for vein pattern databases to improve the search speed and accuracy of identification. In this work, we generate a binary code for each image using texture information. A hierarchical decomposition of Delaunay triangulation based approach for minutiae is proposed and used with binary code to narrow down the search space of the database. Experiments are conducted on two vein pattern databases, and the results show that, while maintaining 100% Hit Rate, the proposed method achieves lower penetration rate than what existing methods achieve.

Classification of Extended Delaunay Triangulation for Fingerprint Indexing

In this paper, we propose an efficient indexing mechanism for the fingerprints based on minutiae triplets. First, we compute the Delaunay triangulation for each fingerprint using the minutiae features. Further, we define a novel representation for the fingerprint named extended triangle set, which is very tolerant to missing minutiae. The defined extended triangle set is an extension of the Delaunay triangulation and is used in combination with a classification approach to narrow down the search space of a fingerprint database. Experiments are conducted on different Fingerprint Verification Competition (FVC) databases and the results show that, while maintaining high hit rate the proposed method achieves lower penetration rate than existing methods.

Hand Vein Authentication System Using Dynamic ROI

This paper presents an efficient authentication system based on hand vein pattern. The stages involved in vein pattern authentication system are image acquisition, Region of Interest (ROI) Extraction, image enhancement, binarization, thinning, feature extraction and matching. We propose an algorithm for extraction of dynamic ROI from the hand vein image. The advantage of dynamic ROI extraction is that, ROI extracted for different hand images varies in size as the size of the hand varies and is possible to extract more features from a larger hand which otherwise is not possible with fixed ROI. A new thinning algorithm is used to extract one pixel thick medial axis vein network from the dynamic ROI and compared the results with matlab’s thinning algorithm. The resulting thinned image may contain some artefacts, and we propose an algorithm to remove these artefacts. The minutiae features that represents the geometric information of the vein pattern is extracted which are bifurcation and ending points. Finally a matching algorithm is applied for authentication. The proposed system is efficient and got the lowest error rate.

Palmprint Recognition Using Fusion of 2D-Gabor and 2D Log-Gabor Features

Palmprint technology is a new branch of biometrics used to identify an individual. Palmprint has rich set of features like palm lines, wrinkles, minutiae points, texture, ridges etc. Several line and texture extraction techniques for palmprint have been extensively studied. This paper presents an intra-modal authentication system based on texture information extracted from the palmprint using the 2D- Gabor and 2D-Log Gabor filters. An individual feature vector is computed for a palmprint using the extracted texture information of each filter type. Performance of the system using two feature types is evaluated individually. Finally, we combine the two feature types using feature level fusion to develop an intra-modal palmprint recognition system. The experiments are evaluated on a standard benchmark database (PolyU Database), and the results shows that significant improvement in terms of recognition accuracy and error rates with the proposed intra-modal recognition system compared to individual representations.

A Score-Based Indexing and Retrieval Technique for Biometric Databases

This paper proposes a score-based indexing technique for biometric databases. A fixed-length index is computed for each image in the database by calculating its match scores against a preselected set of representative images. Further, an efficient storage mechanism (i.e. index space) is developed to arrange the biometric images like traditional database records so that a rapid search is possible. During identification, the retrieval technique finds a list of similar candidates for the query image from the database using voting scheme. Finally, to identify the genuine match from the retrieved similar candidates, we perform a one-one match between the query and each similar candidate using modified Hausdorff distance measure. Experimental results on different databases show a significant performance improvement in terms of response time and identification accuracy compared to the existing indexing methods.

An Efficient Coding Method for Indexing Hand-based Biometric Databases

Biometric identification systems capture biometric (i.e., fingerprint, palm, and iris) images and store them in a central database. During identification, the query biometric image is compared against all images in the central database. Typically, this exhaustive matching process (linear search) works very well for the small databases. However, biometric databases are usually huge and this process increases the response time of the identification system. To address this problem, we present an efficient technique that computes a fixed-length index code for each biometric image. Further, an index table is created based on the indices of all individuals. During identification, a set of candidate images which are similar to the query are retrieved from the index table based on the values of query index using voting scheme that takes a constant time. The technique has been tested on benchmark PolyU palm print database and NTU Vein pattern database. The technique performs with lower penetration rates for 100 % hit rate for both the databases. These results show a better performance in terms of response time and search speed compared to the state-of-the-art indexing methods.

A New Indexing Method for Biometric Databases Using Match Scores and Decision Level Fusion

This paper proposes a new clustering-based indexing technique for large biometric databases. We compute a fixed length index code for each biometric image in the database by computing its similarity against a preselected set of sample images. An efficient clustering algorithm is applied on the database and the representative of each cluster is selected for the sample set. Further, the indices of all individuals are stored in an index table. During retrieval, we calculate the similarity between query image and each of the cluster representative (i.e., query index code) and select the clusters that have similarities to the query image as candidate identities. Further, the candidate identities are also retrieved based on the similarity between index of query image and those of the identities in the index table using voting scheme. Finally, we fuse the candidate identities from clusters as well as index table using decision level fusion. The technique has been tested on benchmark PolyU palm print database consist of 7,752 images and the results show a better performance in terms of response time and search speed compared to the state of art indexing methods.

A New Cluster-Based Indexing Technique for Palmprint Databases Using Scores and Decision-Level Fusion

This chapter proposes a new clustering-based indexing technique for large biometric databases. We compute a fixed-length index code for each biometric image in the database by computing its similarity against a preselected set of sample images. An efficient clustering algorithm is applied on the database and the representative of each cluster is selected for the sample set. Further, the indices of all individuals are stored in an index table. During retrieval, we calculate the similarity between query image and each of the cluster representatives (i.e., query index code) and select the clusters that have similarities to the query image as candidate identities. Further, the candidate identities are also retrieved based on the similarity between index of query image and those of the identities in the index table using voting scheme. Finally, we fuse the candidate identities from clusters as well as index table using decision-level fusion. The technique has been tested on benchmark PolyU palmprint database consisting of 7,752 images and the results show a better performance in terms of response time and search speed compared to the state-of-the-art indexing methods.

Hierarchical Decomposition of Extended Triangulation for Fingerprint Indexing

In biometric identification systems, the identity corresponding to the query image is determined by comparing it against all images in the database. This exhaustive matching process increases the response time and the number of false positives of the system. This chapter presents an efficient indexing algorithm for fingerprint databases to improve the search speed and accuracy of identification. A variant of Delaunay triangulation called extended triangulation is used to make the system robust against distortions. Then the triangles are partitioned into groups such that the retrieval algorithm searches in reduced space of the database. Experiments are conducted on different fingerprint databases, and the results show that while maintaining high hit rate the proposed method achieves lower penetration rate than what existing methods achieve.

Efficient Score-Based Indexing Technique for Fast Palmprint Retrieval

Biometric identification systems capture biometric (i.e., fingerprint, palm, and iris) images and store them in a central database. During identification, the query biometric image is compared against all images in the central database. Typically, this exhaustive matching process (linear search) works very well for the small databases. However, biometric databases are usually huge and this process increases the response time of the identification system. To address this problem, we present an efficient technique that computes a fixed-length index code for each biometric image. Further, an index table is created based on the indices of all individuals. During identification, a set of candidate images which are similar to the query are retrieved from the index table based on the values of query index using voting scheme that takes less time. The technique has been tested on benchmark PolyU palmprint database and the results show a better performance in terms of response time and search speed compared to the state-of-the-art indexing methods.