| Date | Speaker | Title and Abstract |
| Aug 22 | Brendan | Background Subtraction Using Ensembles of Classifiers with an Extended Feature Set The limitations of foreground segmentation in difficult environments using standard color space features often results in poor performance during autonomous tracking. This work presents a new approach for classification of foreground and background pixels in image sequences by employing an ensemble of classifiers, each operating on a different feature type: the three RGB features, the gradient magnitude and orientation features, and eight Haar features. These thirteen features are used in an ensemble classifier where each classifier operates on a single image feature. Each classifier implements a Mixture of Gaussians-based unsupervised background classification algorithm. The non-thresholded, classification decision score of each classifier are fused together by taking the average of their outputs and creating one single hypothesis. The results of using the ensemble classifier on three separate and distinct data sets are compared to using only RGB features through ROC graphs. The extended feature vector outperforms the RGB features on all three data sets, and shows a large scale improvement on two of the three data sets. The two data sets with the greatest improvements are both outdoor data sets with global illumination changes and the other has many local illumination changes. When using the entire feature set, to operate at a 90% true positive rate, the per pixel, false alarm rate is reduced five times in one data set and six times in the other data set. |
| Aug 29 | Soweon | Iris Recognition at a Distance Although iris recognition is one of the most accurate biometric technologies, it has not yet been widely used in practical applications due to user inconvenience during the image acquisition phase. Particularly, users try to place their eye in a small confined region which enables the system to get clear iris image. To overcome the positioning problem, we propose a novel iris image acquisition system based on a pan-tilt-zoom (PTZ) camera and light stripe projection. The PTZ camera can obtain the iris images under position and height variation of users. Instead of capturing only an iris image at a distance, iris regions are extracted from a full face image with high resolution which has less opportunity for losing the iris as well as enables to use both iris images for recognition. Light stripe projection contributes to fast control of the PTZ camera by giving the user's position in real-time. Firstly, the PTZ camera can find a given user's face in the large capture volume by searching 1D vertical line based on the user's horizontal position. Secondly, positions of zoom and focus lens are initialized by the estimated depth between the user's face and the PTZ camera, which result in narrow search range for the optimal focus lens position. Experimental results showed that proposed system captured the iris images which could be used for iris recognition in average time of 2.479 seconds at a distance of 1.5-3m. |
| Sep 5 | Serhat | Incremental Nonnegative Matrix Factorization Nonnegative matrix factorization (NMF) is one of the recent decomposition tools for multivariate data. NMF offers dimension reduction and produces useful representations by converting a data matrix to multiplication of two smaller matrices. The relationship between NMF and clustering has also been revealed and several studies have empirically showed that NMF can be considered as a useful clustering tool. On the other hand, NMF’s batch nature limits its use in some applications. Thus, we have introduced an incremental non-negative matrix factorization (INMF) scheme in order to overcome the difficulties that conventional NMF has in online processing of large data sets. The proposed scheme enables incrementally updating its factors by reflecting influence of each observation on the factorization appropriately. Unlike conventional NMF, with its incremental nature and weighted cost function the INMF scheme successfully utilizes adaptability to dynamic data content changes with a lower computational complexity. Furthermore, we have also extended our method for incremental clustering task. Test results reported for two video applications, namely background modeling in video surveillance and frame based video scene clustering, demonstrate that INMF is capable of online representing data content while reducing dimension significantly. |
| Sep 12 | Alessandra | Mathematical Techniques Applied to the Processing of Electrophoresis Gel Images.
The main purpose of this work is to present efficient ways to automate tasks such as detecting lanes in electrophoresis gel images and unwarping them. The model proposed here consists of correcting the images from distortions, if necessary; removing noise based on mathematical smoothing; correcting and removing the background based on mathematical morphology; and, after these procedures, detecting lanes based on projections. |
| Sep 19 | Jianjiang | Fingerprints - I : Fingerprint acquisition and quality evaluation
An overview of fingerprint recognition system is given. Several fingerprint sensing technologies are then introduced. Fingerprint quality is discussed with an emphasis on NIST fingerprint image quality (NFIQ). |
| Sep 26 | Jianjiang | Fingerprints - II : Feature Extraction in Fingerprints
Fingerprint features are generally categorized into three levels. Level 1 features are ridge orientation, ridge frequency, and singularity. Level 2 features include ridges and minutiae. Level 3 features include pores, ridge contour, dots, incipient ridges, etc. For each type of feature, a typical extraction algorithm is discussed. Finally, some new algorithms for extracting Level 1 features are discussed. |
| Oct 3 | Seminar Cancelled | CSE 40th anniversary! |
| Oct 10 | Jianjiang | Fingerprints - III : Fingerprint Matching
Large intra-class variations and small inter-class distances make robust fingerprint matching a challenging problem. Factors reducing the separability between different fingerprints include internal one (limited number of fingerprint patterns), and (ii) external ones (poor image quality, small overlapped area, unknown alignment and nonlinear distortion). In this seminar, we talk about the problems in fingerprint matching, state-of-the-art in fingerprint matching, and a descriptor-based minutiae matching algorithm. |
| Oct 17 | Seminar substituted with Dr. Shipeng Yu's talk. | Theory and Applications of Bayesian Co-Training |
| Oct 24 | Abhishek | Secure Biometric Recognition
Reliable identity management is crucial in today's world, especially in view of illegal transactions of the order of US$50 billion a year. Biometrics offers a promising solution to the current identity crisis situation but technology is still not mature enough for large scale deployments. In this talk we will discuss various vulnerabilities of a common biometric recognition system and what additional properties need to be ascertained by a biometric recognition system in order to get ready for large scale deployments. Specifically we will focus on intrinsic limitations of a biometric system and enlist certain adversary attacks that can be staged. Also various requirements of a secure biometric recognition will be discussed. |
| Oct 31 | Abhishek | Security in Biometrics - II
In continuation with the last seminar where we discussed some major vulnerabilities and consequences of overlooking such vulnerabilities in a biometric recognition system, in this seminar we shall focus on template security. We will identify two main categories of template security approaches: Feature transformation based and Biometric Cryptosystem, and discuss some examples of each of those. Further, we shall describe a hybrid scheme that benefits from both the aforementioned approaches. We shall also try to identify some advantages of multi-biometric systems towards building a secure recognition system. At last we shall discuss some major challenges that still face the construction of a practically secure biometric recognition system. |
| Nov 7 | Unsang | Face Biometrics - I
Face has been studied over the past three decades for the purpose of automatic person recognition in 2D, 3D and video domain. Its applications range from law enforcement, and civil applications, to surveillance systems. There has been substantial improvement in face recognition performance since its conception, but it is still far lower than acceptable for use in most applications. The difficulties in face recognition arise from the large intra-class variability due to pose, lighting, expression, and age variations. This talk will review the some of the basic and state-of-the-art techniques in face detection and face recognition. |
| Nov 14 | Unsang | Face Biometrics - II
Face has been studied over the past three decades for the purpose of automatic person recognition in 2D, 3D and video domain. Its applications range from law enforcement, and civil applications, to surveillance systems. There has been substantial improvement in face recognition performance since its conception, but it is still far lower than acceptable for use in most applications. The difficulties in face recognition arise from the large intra-class variability due to pose, lighting, expression, and age variations. This talk will review the some of the basic and state-of-the-art techniques in face detection and face recognition. |
| Nov 21 | Guest Speaker: Dr. Kevin W. Bowyer, University of Notre Dame |
Next-Generation Iris Biometrics This talk begins with an overview of the major approach to iris biometrics. This approaches uses a binary "iris code" that represents the texture pattern of an individual eye. The importance of masking inconsistent bits in the code is explained and illustrated as a means to improved accuracy. The importance of incorporating knowledge of the degree of pupil dilation is also explained and illustrated. Finally, problems that arise due to the wearing of contact lenses are illustrated as an example current research problem. |
| Nov 28 | University Closed | |
| Dec 5 | Pavan | Semi-supervised Learning - I |
| Dec 12 | Pavan | Semi-supervised Learning - II |
| Dec 19 | Jung-Eun | Content based image retrieval -I |
| Dec 26 | Jung-Eun | Content based image retrieval -II |