Tayfun Akgul

Suspect identification based on descriptive facial attributes

We present a method for using human describable face attributes to perform face identification in criminal investigations. To enable this approach, a set of 46 facial attributes were carefully defined with the goal of capturing all describable and persistent facial features. Using crowd sourced labor, a large corpus of face images were manually annotated with the proposed attributes. In turn, we train an automated attribute extraction algorithm to encode target repositories with the attribute information. Attribute extraction is performed using localized face components to improve the extraction accuracy. Experiments are conducted to compare the use of attribute feature information, derived from crowd workers, to face sketch information, drawn by expert artists. In addition to removing the dependence on expert artists, the proposed method complements sketchbased face recognition by allowing investigators to immediately search face repositories without the time delay that is incurred due to sketch generation.

Towards automated caricature recognition

This paper addresses the problem of identifying a subject from a caricature. A caricature is a facial sketch of a subjects face that exaggerates identifiable facial features beyond realism, while still conveying his identity. To enable this task, we propose a set of qualitative facial features that encodes the appearance of both caricatures and photographs. We utilized crowdsourcing, through Amazons Mechanical Turk service, to assist in the labeling of the qualitative features. Using these features, we combine logistic regression, multiple kernel learning, and support vector machines to generate a similarity score between a caricature and a facial photograph. Experiments are conducted on a dataset of 196 pairs of caricatures and photographs, which we have made publicly available. Through the development of novel feature representations and matching algorithms, this research seeks to help leverage the ability of humans to recognize caricatures to improve automatic face recognition methods.

Experimental analysis of detection and localization of multiple emitters in multipath environments

An effective way for the joint detection and localization of multiple RF transmitters in a multipath environment is to enumerate the number of paths using the minimum-description-length information-theoretic algorithm, and to then measure the angle of arrival of each path, using an antenna array with a high-resolution direction-finding algorithm, such as MUSIC. The possible propagation paths are the angles corresponding to the peaks of the MUSIC pseudo-spectrum. Since more than one path may correspond to a single emitter source, further processing is required. The time-domain signals incident from these paths are then extracted with beamforming techniques, such as minimum variance, in order to estimate their coefficients of correlation with each other. These correlation coefficients are used to decide whether or not these paths correspond to the same emitter. Among the paths that appear to originate from the same source, the path with the time signal that contains the highest power is selected as the original path of the source. Hence, the number of emitters and their angles of arrival are jointly estimated. A performance analysis of the method is presented via real-time laboratory experimentation and discussed in this paper. To demonstrate the effectiveness of the proposed technique, experimental results with two uncorrelated sources were compared to experimental results with a single source and a reflector. All of the experiments were conducted in an anechoic test chamber.

Estimation of Spectral Exponent Parameter of Process in Additive White Background Noise

An extension to the wavelet-based method for the estimation of the spectral exponent, , in a process and in the presence of additive white noise is proposed. The approach is based on eliminating the effect of white noise by a simple difference operation constructed on the wavelet spectrum. The parameter is estimated as the slope of a linear function. It is shown by simulations that the proposed method gives reliable results. Global positioning system (GPS) time-series noise is analyzed and the results provide experimental verification of the proposed method.

Can an Algorithm Recognize Montage Portraits as Human Faces? [In the Spotlight]

The face is one of the most important features of a human being, and its recognition is essential. Humans have to know immediately who they are facing-an enemy or a friend? We are most likely genetically coded not only to recognize faces, but also to extract the characteristics of expressions of faces for survival. The need to recognize faces probably accelerated the evolution of human intelligence. Now, due to the significant leaps that computing power took over the past decade, the time has come for machines/computers to mimic the same process-the ability to achieve reliable face recognition as successful as human beings or even better. Face recognition is, of course, a fertile ground for the development of new algorithms and applications in a myriad of fields. For example, CNN recently reported [1] that Facebook was testing a new feature that uses face recognition to help in the tagging of photos. For over a decade, there has been an increasing interest in face recognition in diverse fields such as pattern recognition, computer vision, telecommunications, video, security and Internet applications, and cognitive psychology. Among various face recognition algorithms developed, they are mainly classified into two groups: pose dependent and pose invariant [2]. Pose-dependent algorithms rely on twodimensional (2-D) images of different poses of faces, while pose-invariant techniques are based on three-dimensional (3-D) models. Here, without discussing the details of such approaches, let us ask a simple question: How does an artist or a caricaturist capture the characteristics of faces and with simple line drawings makes us successfully recognize faces, in many cases better than the full image of a person? If the answer can be found, we may come up with better face recognition methods.

EMD-based analysis and denoising of GPS data

In this study Empirical Mode Decomposition (EMD)-based denoising of the Global Positioning System (GPS) data is explored. EMD is a data-driven new technique which is proposed by Huang and improved by Flandrin and his colleagues. For only white noise or only colored noise cases, a denoising algorithm is proposed in literature. This study expands the denoising scheme for white + colored noise case, which is applicable to some real signals such as Global Positioning System (GPS) based geodetic time series. The results for synthetic data and real GPS data are presented.

Matching caricatures to photographs

Facial caricatures are informative funny images that allow us to identify a subject even with a few lines and dots. Matching caricatures to photographs is a challenging cross-modal face matching problem. This paper addresses this problem by defining a set of qualitative face features both for caricatures and photographs where features are automatically extracted from photos and manually labeled in caricatures. Additionally we release a publicly available caricature-photograph database with 200 caricatures and corresponding photomates. In our experiments, we use genetic algorithms and logistic regression and achieve over

Principal Component Analysis of the Fractional Brownian Motion for 0 l H l 0.5

\mathrm{30.0}\,\%