Seungsin Lee

Imaging for concealed weapon detection: a tutorial overview of development in imaging sensors and processing

Manual screening procedures for detecting concealed weapons such as handguns, knives, and explosives are common in controlled access settings like airports, entrances to sensitive buildings, and public events. The detection of weapons concealed underneath a persons clothing is an important obstacle to the improvement of the security of the general public as well as the safety of public assets like airports and buildings. It is desirable to be able to detect concealed weapons from a standoff distance, especially when it is impossible to arrange the flow of people through a controlled procedure. The goal is the eventual deployment of automatic detection and recognition of concealed weapons. It is a technological challenge that requires innovative solutions in sensor technologies and image processing. A number of sensors based on different phenomenology as well as image processing support are being developed to observe objects underneath peoples clothing. The main aim of this article is to provide a tutorial overview of these developments.

Discrete-time models for statistically self-similar signals

Wide-sense statistical self-similarity in continuous-time random processes is defined through invariance of its first-order and second-order statistics to scaling in time. Since scaling has an unambiguous definition in continuous-time but not in discrete-time, researchers have provided various definitions of discrete-time self-similarity without reference to scaling. This paper proposes a discrete-time continuous-dilation scaling operator and develops a framework based on it for formulating statistical self-similarity from first principles in a manner analogous to the continuous-time development. Relationship between the resulting model and fractional order transfer function systems is presented. The potential for using this model in applications involving long-range dependent phenomena is explored.

A Video Processing Approach for Distance Estimation

The paper presents a passive ranging method for estimating distances to an object using a video sequence gathered from a moving platform. The motivation is provided by potential application to object distance estimation using video data from general aviation aircraft. The method exploits scale changes of an object in the video sequence, as inferred by processing wavelet transforms of video frames, to compute distance. The underlying principles are presented along with results of bench experiments

Noise reduction and object enhancement in passive millimeter wave concealed weapon detection

Passive MM wave offers the advantage of penetration for concealed weapon detection. Its ability to penetrate through fog, smoke, clothing etc. makes it an attractive candidate to look for weapons concealed underneath a persons clothing. The sensor technology has advanced to a point where it is possible to generate real time video sequences. However, noise and blur are still severe problems. The work reported here investigates the problem of simultaneous noise suppression and object enhancement in passive millimeter wave video sequences. The basis for the approach is provided by undecimated wavelet transforms in the spatial dimension and motion compensated filtering in the temporal dimension. The paper presents the underlying principles of the approach as well as experimental results.

Scale-based formulations of statistical self-similarity in images

Statistically self-similar images that are segments of two dimensional self-similar random fields, have been useful in the analysis and synthesis of certain types of textures. Whereas a rigorous definition of self-similarity in continuous-space is based on spatial scaling, current treatments in digital image processing are based on ad-hoc approaches rather than on spatial scaling mainly because of the unavailability of continuous scaling in discrete-space. This paper presents a formulation based on such a continuous scaling operator leading to a more general and versatile characterization of statistical self-similarity in images.

Signal processing models for discrete-time self-similar and multifractal processes

The paper discusses and presents results from recent investigation into three problems. (A) Approaches have previously been developed for describing self-similarity in discrete-time random processes using a discrete-time continuous dilation operator. It is shown here that processes self-similar under this construct, called a discrete-time self-similar system (DTSS), are related to prior discrete-time constructs; specifically they can generate asymptotically second order self-similar processes. (B) The advantage of using long-range prediction of long-range dependent processes is tested. It is shown that for combined long-range and short-range prediction for tracking with a sequence that has bounded increments, long range prediction does not offer a significant tracking advantage

Characterization of linear scale invariant system input/output relationships and synthesis of network traffic traces

It has been shown that discrete-time self-similar processes can be synthesized by the discrete-time linear scale invariant system (DLSI) models proposed by Zhao and Rao. This paper reports results of subsequent studies on the DLSI systems to characterize properties such as long-range dependency. This paper also shows that self-similar processes generated by these systems have fractional Gaussian characteristics regardless of the marginal distribution of the white noise input. This property can be verified by comparing the autocovariance functions of the fractional Gaussian noise and the system outputs. It is also shown that the system is able to generate statistically self-similar signals whose parameters coincide with those observed in network traffic traces.

Algorithms for scene restoration and visibility estimation from aerosol scatter impaired images

Attenuation and backscatter from aerosol suspension in the atmosphere cause drop in visibility which manifests as haziness or fogginess in images captured of objects at a distance, especially from airborne cameras. Two iterative algorithms, based on a physical model relating image intensity to ground reflectance as a function of the scatter process, are developed to restore the images and estimate the scatter coefficient. The algorithms are guaranteed to converge to unique solutions. Examples are provided with synthetic and real data.

Discrete-time scale invariant systems: Relation to long-range dependence and farima models

A discrete-time version of linear scale-invariant systems was provided by Zhao and Rao. This formulation, called a DLSI system, was derived using a continuous dilation operator in discrete-time as a direct analog of the continuous-time linear scale-invariant system formulation of Wornell. While simulations had shown that DLSI systems were capable of generating self-similar data such as those found in network traffic, answers to questions regarding their relationship to other models had not been found. This paper investigates such relationships. It derives results establishing parameter ranges for DLSI systems that give rise to long-range dependent outputs for white noise inputs. Furthermore, it shows that the basic DLSI system is expressible as a special fractional ARIMA model. Finally, the DLSI model is applied to scene classification in variable bit rate (VBR) MPEG video to provide a preliminary indication of potential application.Multiple antenna systems that operate at high data rates require simple yet effective space-time transmission schemes to handle the large traffic volume in real time. Recently, different block space-time codes, such as: Orthogonal Designs, VBLAST, LD (Linear Dispersion) codes, have been introduced. Due to the high complexity of the optimum ML (Maximum Likelihood) detection, sub-optimal detection methods like sequentially nulling and canceling for the VBLAST, have been proposed for non orthogonal block space-time codes. A new lattice decoder has been proposed using sphere decoding, but this method yet suffers from complexity and sensitivity to choose the radius of the sphere. We present a new sphere decoding algorithm that has lower complexity than other schemes and performs very close to ML decoding.

Synthesis models for N-dimensional discrete-space self-similar signals

New formulations and models are proposed for describing statistical self-similarity in general N-dimensional settings. By using a matrix scaling operator for defining statistical self-similarity, a wide class of continuous-space N-D processes can be characterized as self-similar with respect to specific matrix classes. Further, discrete-space versions of N-D statistical self-similarity are treated through a discrete-domain scaling operation. The mathematical basis for the approaches is provided along with 2-D synthesis examples