8.1 Dimensionality of Index Key Vector

In biometric data indexing, an important judgment is about the dimensionality of the index key vector. The dimensionality of the index vector should be in such a manner that it reduces the search space significantly and at the same time retrieves a minimum number of candidates without compromise in accuracy. In the following, how this objective is achieved with respect to the different biometric data are summarized.

Majority of the iris-based indexing techniques consider the creation of index key from iris texture pattern [21, 22, 18, 19]. Among these texture based methods, SIFT feature based method [19] gives the better result to the best of knowledge. This method uses 128-dimensional index key vector. In contrast to the existing practices, this is the first time iris biometric based indexing mechanism is proposed using Gabor energy features of iris texture. The Gabor energy features allow to derive only 12-dimensional index key for iris biometric. Indeed this is quite low dimension of index key vector in comparison to the several existing work. The summary of iris feature representation is shown in Table 8.1.

From the study of existing literature, it is observed that minutiae based fingerprint indexing uses low dimensional index key vector for fingerprint data indexing. Moreover, the MCC based fingerprint indexing [4] is treated as the most efficient minutiae based indexing technique. This technique utilizes at least 15 dimensional index key vector. On the other hand, the feature extraction technique using two closest point triangulation method generates 8-dimensional index key vectors. To add more, the scale and rotation invariant properties of index keys make the proposed indexing scheme more robust compared to other higher dimensional index key vector reported elsewhere. Table 8.1 shows the summary of fingerprint features.

To achieve the improved accuracy of the indexing system, the existing face biometric data indexing techniques advocate high dimensional index key vector [20, 14, 16]. According to the reported work, it is observed that the method proposed by Kaushik et al. [14] gives better accuracy. This technique uses 128-dimensional index key vectors. In contrast, the proposed face biometric data indexing mechanism considers 69-dimensional index key vectors. It may be noted that in the proposed 69-dimensional face indexing mechanism, the first four dimensions of the index key vector are constituted from the key point

information, which are used to index the face data. The next 64 dimensions, which contain the SURF feature descriptors information, are used to match the face template. The last dimension of the index key vector is the identity of a subject. The summary of face feature are reported in Table 8.1.

To the best of exhaustive literature survey on multimodal biometric data indexing, only a few number of research work have been reported. Out of these reported work, the approach proposed in [9] performs better than the other approaches. However, this approach uses 256-dimensional index key vector to index the multimodal biometric data of face and fingerprint traits. On the other hand, the proposed approach generates 4-dimensional index key vector from the relative match scores (see Table 8.1). Further, the relative scores are calculated against a set of reference subjects corresponding to different unimodal traits as well as multimodal traits. It may please be noted that three biometric traits: iris, fingerprint and face biometric traits are considered and combined the scores using SVM-based score level fusion technique which enable to solve the problem with such a low-dimensional index key vector.

8.2 Storing and Retrieving

Storage structure is another important issue in any indexing mechanism. It may be noted that storage structure should vary depending on the index key and the data structure also highly influences the retrieving efficiency. Further, indexing technique needs extra memory overhead. The investigation to achieve the best storage structure for the indexing of iris, fingerprint, face and multimodal biometric data are discussed in the following.

Existing iris data indexing techniques use different storing structures which are more relevant to the traditional retrieval techniques [22, 24, 13]. The tree-based storage structure is commonly used to store the iris data. Although, the retrieving efficiency in tree-based storing may not be acceptable for a large number of entries into the database. On the contrary, table-based storage structure is used in the proposed iris indexing technique. A table is maintained for a dimension of the iris index key vector and stored all iris data into the table based on the value of that dimension in the index space. This proposed storage structure helps to retrieve a small set of candidates from the database in constant time using the proposed retrieving mechanism. Further, the low dimensionality of the index key vector reduces the memory overhead for the iris indexing system.

From the current literature, it can be seen that majority of the fingerprint indexing approaches [2, 3, 8] follow continuous classification mechanism and for this purpose they use linear storage structure. On the other hand, to store the fingerprint data, three different storing structures: linear, clustered and clustered kd-tree are used. In the fingerprint indexing approach, k-means clustering technique is applied on index keys to cluster the fingerprint data. Three search techniques are experimented for these three storing structures. It is observed that clustered and clustered kd-tree searches efficiently reduce the search space of fingerprint data. The proposed approach deals with low dimensional index key vector, hence, less indexing memory overhead is required. Nevertheless, the clustered kd-tree based storage requires an extra memory overhead to store the cluster and kd-tree information.

Existing face indexing techniques hardly explore the storing structures of the face-based indexed data [20, 14]. In the proposed face indexing, biometric data are stored into a two level index space in the database. In the first level, the face data are divided into two groups based on the first dimension of the index keys and in second level, a 3-dimensional index cube is created based on the next three dimensions of the index keys. A linear or kd-tree structure is used within a cell of an index cube to store the face data. A hash function is applied on the index keys to store and retrieve the face data. The proposed storing and retrieving methods allow to efficiently generate an accurate candidate set with similar template for a given query. However, the proposed storing technique needs a small memory to store the two level index space into the database as only four dimensions of the index key vector are used to create the index space.

Existing work on multimodal biometric data indexing use linear or kd-tree storage structures and follow linear and kd-tree based searching [9, 12]. On the other hand, table-based storage structure is used to store the identities of multimodal biometric data and follow hashing technique to retrieve the identities from the database. The subject identities are stored into the database based on the feature values corresponding to different traits in index key vector and retrieved a candidate set for each biometric trait. Further, rank level fusion technique is applied by applying SVM rank to combine the retrieved candidates. As the multiple entries of a subject identity are stored, the proposed approach requires a bit extra memory for this purpose. As a leverage of the extra memory overhead, the proposed technique sufficiently narrow downs the search space and precisely retrieves the most similar templates.

8.3 Performance of Indexing Techniques

The performance of a biometric based identification systems relies on the efficiency and accuracy of the indexing technique. The experimental results of the proposed indexing mechanisms for unimodal and multimodal biometric traits substantiate the efficiency and accuracy of the proposed methods. In Table 8.2, the performances of different indexing techniques exercised in this research are compared. The outcomes of the proposed indexing techniques for iris, fingerprint, face and multimodal biometric traits are summarized in the following.

The index space organization of iris data allows to retrieve iris data from the database in constant time without compromising the accuracy. An exhaustive evaluation has been done with different iris databases. On the average 82.79% rank one HR and 13.78% PR can be achieved for all iris databases (see Table 8.2). The proposed approach is also capable to achieve 96.75% CMS on an average at the 30^th rank as shown in Fig. 8.1. The method has also been shown to perform better than the best of the existing approaches [19]. The experimental results indicate that the iris biometric data indexing approach can be applied to any real time iris biometric-based identification system which deals with a large number of iris biometric data.

The proposed fingerprint-based indexing approach ables to achieve a higher accuracy with the proposed low dimensional index key. An extensive study has been performed with different fingerprint databases and the average results of all fingerprint databases in clustered kd-tree based indexing are reported in Table 8.2 which indicates that fingerprint indexing can achieve on the average 83.83% rank one HR and 14.05% PR and retrieve a set of similar fingerprint templates in the order of milliseconds. From Fig. 8.1, it can be seen that the approach is able to achieve 99.83% CMS at the 30^th rank. Further, better accuracy can be achieved in linear or cluster based fingerprint data indexing though the retrieving time is higher than clustered kd-tree based indexing technique. Depending the applications’ requirement, any one of these indexing technique can be utilized. It may be concluded that the proposed clustered kd-tree based indexing technique outperforms the existing techniques.

Table 8.2: Performances of the different indexing approaches

Figure 8.1: CMC curves of different indexing approaches.

Experimental results on different face databases prove that the proposed face indexing approach provides better results than the existing techniques [14, 16, 20]. The average performance on different face databases is reported in Table 8.2. On the average 93.52% rank one HR and 9.30% PR can be achieved. The proposed approach, in fact, performs the face data retrieval in the order of millisecond. Also, the approach can achieve on the average 96.69% CMS (see Fig. 8.1). Moreover, the experimental results establish the potential of the proposed approach for handling a large face database of a face biometric based identification system.

Comprehensive evaluation on a virtual user database shows the effectiveness of the proposed multimodal biometric data indexing approach. The proposed multimodal biometric indexing approach can achieve 96.11% rank one HR and 13.86% PR and retrieves a candidate set in constant time (see Table 8.2). A 99.25% CMS at the 30^th rank (see Fig. 8.1) can also be achieved. It is observed that the multimodal indexing approach performs better than the existing approaches [9, 12]. It is evident from the experimental results that the proposed multimodal biometric indexing approach can be applied to any large scale application. Further, the proposed approach can handle any number of biometric traits for a multimodal biometric based identification system.

8.4 Threats to Validity

The experiments of proposed indexing mechanisms are involved with different biometric databases and different parameters. Eventually, the experimental results reported in this book are subjected to the validity of the available resources and assumptions on values of parameters. In the following, the validity of the experiments and experimental results are discussed.

8.5 Future Scope of Work

While this work have made significant improvement in the development of biometric data indexing methods that facilitate the design of fast and reliable biometric identification systems, it opens up several future avenues for research. Some of them are mentioned below.

• Most of the algorithms for different biometric data indexing, including proposed techniques, are studied with the biometric databases whose sizes are in the order of thousands. There is no such large database in the order of millions for the research community. So, creating a large biometric database with different biometric traits is a promising topic of research.
• The performance of the biometric data indexing algorithms degrade when the quality of the captured biometric traits is low. Hence, there is a scope for research to handle the low quality biometric traits for indexing purpose.
• In this book, the biometric images are used which are captured from cooperative users. That means users cooperate with the system when the biometric samples are taken from the users. Biometric data indexing with non-cooperative users’ data is a challenging research area.
• The Gabor energy features are investigated in the proposed iris data indexing technique. In future, other texture feature extraction techniques for iris biometric can be explored to improve the accuracy of iris data indexing.
• A well-known limitation of the k-means clustering is that an inappropriate choice of k may yield poor results. In the proposed fingerprint data indexing approach, k-means clustering technique is used to group the fingerprint data. Other clustering techniques can be investigated in future research.
• Apart from the geometry of the minutiae triplets and texture pattern around the minutiae point, many minutiae-based fingerprint matchers use additional attributes like minutia type, ridge counts, ridge curvature, ridge density and local texture features to achieve high recognition rates. These attributes can also be incorporated into the fingerprint-based indexing in future study.
• Face biometric data indexing is proposed for frontal face image using SURF features. However, SURF method may not good for side profile or occluded face images to extract features. So, these types of images demand great attention for further research.
• In the proposed multimodal biometric data indexing, SVM-based score fusion technique is explored. Though, to the best of knowledge this technique performs better than other reported techniques till date. On the other hand, accuracy of SVM-based technique rely on the training data. Hence, further research can possible to improve the accuracy of the score fusion technique.
• In this book, three biometric traits are used for indexing. However, there are several other biometric traits used in different applications. The indexing with the biometric traits other than iris, fingerprint and face are yet to be addressed.
• Security of the biometric templates is another concern in recent days. To prevent the theft of biometric data, cancelable templates are generated from the different biometric traits. Generating cancelable biometric index key for the security of biometric data would be thought as a new area of research.
• Finally, a formal analysis for cost-benefit of a biometric indexing system based on parameters such as performance gain (HR, PR, CMS), speed up, physical cost of the system and security needs to be developed in order to enable biometric system developers to rapidly design an indexing system that is most appropriate for the application on hand.

Bibliography

• [1] H. Bay, A. Ess, T. Tuytelaars, and L. V. Gool. SURF: Speeded Up Robust Features. Computer Vision and Image Understanding , 110(3):346–359, 2008. doi: http://dx.doi.org/10.1016/j.cviu. 2007.09.014.
• [2] B. Bhanu and X. Tan. Fingerprint Indexing Bsed on Novel Features of Minutiae Triplets. IEEE Transactions on Pattern Analysis and Machine Intelligence, 25(5):616–622, 2003. ISSN 0162-8828.
• [3] R. Cappelli. Fast and Accurate Fingerprint Indexing Based on Ridge Orientation and Frequency. IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics, 41(6):1511–1521, 2011.
• [4] R. Cappelli, M. Ferrara, and D. Maltoni. Fingerprint Indexing Based on Minutia Cylinder-Code. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(5):1051–1057, 2011.
• [5] J. Daugman. The Importance of being Random: Statistical Principles of Iris Recognition. Pattern Recognition, 36(2):279–291, 2003.
• [6] J. Daugman. How Iris Recognition Works. IEEE Transactions on Circuits and Systems for Video Technology, 14(1):21–30, 2004.
• [7] G. Du, F. Su, and A. Cai. Face Recognition using SURF Features. In Proceedings of the SPIE Pattern Recognition and Computer Vision (MIPPR 2009), volume SPIE-7496, pages 749628–1–749628–7, Yichang, China, October-November 2009.
• [8] A. Gyaourova and A. Ross. A Novel Coding Scheme for Indexing Fingerprint Patterns. In Proceedings of the 7th International Workshop Statistical Pattern Recognition, pages 755–764, Orlando, USA, December 2008.
• [9] Aglika Gyaourova and Arun Ross. Index Codes for Multibiometric Pattern Retrieval. IEEE Transactions on Information Forensics and Security, 7(2):518–529, 2012.
• [10] L. Hong, Y. Wan, and A. Jain. Fingerprint Image Enhancement: Algorithm and Performance Evaluation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 20(8):777–789, 1998. ISSN 0162-8828.
• [11] A. K. Jain, S. Prabhakar, L. Hong, and S. Pankanti. Filterbank-based Fingerprint Matching. IEEE Transactions on Image Processing , 9(5):846–859, 2000.
• [12] U. Jayaraman, S. Prakash, and P. Gupta. An Efficient Technique for Indexing Multimodal Biometric Databases. International Journal of Biometrics (IJBM), 1(4):418–441, 2009.
• [13] U. Jayaraman, S. Prakash, and P. Gupta. An Efficient Color and Texture based Iris Image Retrieval Technique. Expert Systems with Applications, 39(5):4915–4926, 2012.
• [14] V. D. Kaushik, A. K. Gupta, U. Jayaraman, and P. Gupta. Modified Geometric Hashing for Face Database Indexing. In Proceedings of the 7th international conference on Advanced Intelligent Computing Theories and Applications: with Aspects of Artificial Intelligence, pages 608–613, Zhengzhou, China, August 2011.
• [15] B. A. Kitchenham, S. L. Pfleeger, D. C. Hoaglin, and J. Rosenberg. Preliminary Guidelines for Empirical Research in Software Engineering. IEEE Transactions on Software Engineering, 28(8): 721–734, 2002.
• [16] K. H. Lin, K. M. Lam, X. Xie, and W. C. Siu. An Efficient Human Face Indexing Scheme using Eigenfaces. In Proceedings of the IEEE International Conference on Neural Networks and Signal Processing, pages 920–923, Nanjing, China, December 2003.
• [17] B. S. Manjunath and W. Y. Ma. Texture Features for Browsing and Retrieval of Image Data. IEEE Transactions on Pattern Analysis and Machine Intelligence, 18(8):837–842, 1996. doi: http://dx.doi.org/10.1109/34.531803.
• [18] H. Mehrotra, G. S. Badrinath, B. Majhi, and P. Gupta. Indexing Iris Biometric Database using Energy Histogram of DCT Subbands. In Proceedings of the International Conference on Contemporary Computing, volume CCIS-40, pages 194–204, Noida, India, August 2009. doi: http://dx.doi.org/10.1007/978-3-642-03547-0_19 .
• [19] H. Mehrotraa, B. Majhia, and P. Gupta. Robust Iris Indexing Scheme using Geometric Hashing of SIFT Keypoints. Journal of Network and Computer Applications, 33(3):300–313, 2010. doi: http://dx.doi.org/10.1016/j.jnca.2009.12.005.
• [20] P. Mohanty, S. Sarkar, R. Kasturi, and P. J. Phillips. Subspace Approximation of Face Recognition Algorithms: An Empirical Study. IEEE Transactions on Information Forensics and Security , 3(4):734–748, 2008.
• [21] R. Mukherjee. Indexing Techniques for Fingerprint and Iris Databases. Master’s thesis, College of Engineering and Mineral Resources at West Virginia University, 2007.
• [22] R. Mukherjee and A. Ross. Indexing Iris Images. In Proceedings of the 19th International Conference on Pattern Recognition , pages 1–4, Tampa, Florida, USA, December 2008. ISBN 9781424421756. doi: http://dx.doi.org/10.1109/ICPR. 2008.4761880.
• [23] N. B. Puhan and N. Sudha. A Novel Iris Database Indexing Method using the Iris Color. In Proceedings of the 3rd IEEE Conference on Industrial Electronics and Applications (ICIEA 2008), pages 1886–1891, Singapore, June 2008. doi: http://dx.doi.org/10.1109/ICIEA.2008.4582847 .
• [24] M. Vatsa, R. Singh, and A. Noore. Improving Iris Recognition Performance using Segmentation, Quality Enhancement, Match Score Fusion, and Indexing. IEEE Transactions on Systems, Man, and Cybernetics, Part B, 38(4):1021–1035, 2008.