Recovering the Frontal Facial Image from Surveillance Video
Besma Abidi (UTK) and Arun Ross (WVU)
Quality Based Restitution of Iris Features in High Zoom Images for Less Constrained Iris Recognition System
Stephanie Schuckers (Clarkson), Natalia Schmid, Aditya Abhyankar and Lawrence Hornak (WVU)
Automatic High Resolution Retrieval of Tattoos for Victim and Suspect Identification
Anil K. Jain (MSU)
Securing Multibiometric Templates Using Fuzzy Vault
Anil K. Jain (MSU) and Arun Ross (WVU)
Indexing Large‐scale Multimodal Biometric Databases
Anil K. Jain (MSU) and Arun Ross (WVU)
Summaries:
Recovering the Frontal Facial Image from Surveillance Video
Besma Abidi (UTK) and Arun Ross (WVU)
The objective of this proposal is to construct an optimal frontal view from a series of frames obtained from surveillance video. Two techniques (2D and 3D) will be implemented and compared in terms of real time requirements and performance of a recognition engine. The study will be conducted on a variety of sample videos to mimic various real life scenarios and performance evaluated using existing surveillance databases.
Quality Based Restitution of Iris Features in High Zoom Images for Less Constrained Iris Recognition System
Stephanie Schuckers (Clarkson), Natalia Schmid, Aditya Abhyankar and Lawrence Hornak (WVU)
This necessity to have short range eye scanner distance poses a serious limitation in terms of their usability. “Iris recognition from distance’ has received limited attention in literature and is a very challenging problem for the following reason: (1) Lang range distance degrades the overall iris quality. (2) The effect of various noise elements like motion blur, angular deformation etc. gets amplified with the distance. (3) As distance increases, iris recognition techniques for restoration of important iris features from high resolution iris images for efficient long range iris recognition. Following issues will be studied: (1) Design of eye‐scanner distance base adaptive quality metric. (2) Study and analysis of various iris features at different resolutions. (3) Adaptive quality metric based iris segmentation and encoding methodologies. (4) Development of reliable way of formulating distant iris templates using quality restitution of iris features for more dependable recognition with fewer constraints.
Automatic High Resolution Retrieval of Tattoos for Victim and Suspect Identification
Anil K. Jain (MSU)
Tattoos are imprints on the skin useful for identifying the non‐skeletalized body of a victim, or a suspect using a false identity. Tattoos also serve as an indicator of social status, personality, religious affiliations, or criminal organization affiliation of individuals. A wide cross section of the population bears tattoos, from fashion models to known criminals and gang members. Various law enforcement agencies maintain a database of scars, mark & tattoos for the purpose of victim and suspect identification. For this reason, ‘ANSI/NIST – Data Format for the Interchange of Fingerprint, Facial, & Scar Mark & Tattoo Information’ was released to ensure uniformity in the capture and exchange of tattoo data. This protocol utilizes semantic (category) labels (viz., “human forms and features”, “animals”, “plants”, etc.) to characterize each tattoo entry during data collection; the retrieval is, therefore, primarily based on textual queries. This matching and retrieval method is not only time consuming, but lacks objectivity. Image or pattern‐based retrieval, on the other hand, is more appropriate since it lends itself to automated queries based on the content of visual imagery or pattern. We propose to design and build a prototype tattoo matching and retrieval system based on image content and semantic categories.
Securing Multibiometric Templates Using Fuzzy Vault
Anil K. Jain (MSU) and Arun Ross (WVU)
Template security is critical in biometric system design because stolen biometric templates, unlike passwords, cannot be revoked. A number of approaches, including encryption, watermarking, fuzzy vault and revocable template have been proposed to secure templates. However, these approaches have been proposed primarily to secure a single template. A multibiometric system requires the storage of several templates for the same user corresponding to different biometric sources. Therefore, template security is even more critical in multibiometric systems. While it is possible to apply the template protection schemes individually to these templates and combine the authentication results at the decision level, such an approach is not optimal in terms of accuracy and security. We propose a unified scheme to secure multiple templates by (i) transforming features from different biometric sources (e.g., fingerprint minutiae and iriscodes) into a common representation, (ii) performing feature‐level fusion to improve recognition accuracy, and (iii) constructing a single fuzzy vault to secure the fused multibiometric template. We will develop a fully automatic implementation of a multibiometric fuzzy vault that can handle different scenarios such as multiple samples (e.g., two impressions from the same finger), multiple instances (e.g., left and right irises) and multiple traits (e.g., fingerprint, face and iris). We will demonstrate the performance of the proposed multibiometric fuzzy vault in terms of its accuracy and security on public domain (WVU, FVC, CASIA and XMVT) databases.
Indexing Large‐scale Multimodal Biometric Databases
Anil K. Jain (MSU) and Arun Ross (WVU)
Efficient retrieval of pertinent identities from a multimodal biometric database is a challenging task due to the large number of enrolled subjects and the availability of multiple biometric traits corresponding to each subject. Identification systems requiring a short response time will, therefore, be at a disadvantage when searching through the entire database to determine the identity of an individual. This project will explore the design of indexing (or filtering or binning) schemes to define an efficient search and retrieval strategy in multimodal biometric databases. Given the input biometric data of an individual, the goal of indexing is to reduce the search space of possible identities by appropriately partitioning the target biometric database. Further, indexing is likely to increase the overall matching accuracy. While such schemes exist for individual modalities such as fingerprints, multimodal indexing is a yet unexplored problem and has the potential to facilitate rapid and accurate search operations in large‐scale multimodal databases.