We have studied the use of point spread function (PSF) calculated by the wave optics for restoration of out-of- focused images. According to the theory, PSF and its Fourier transform OTF depend only on the defocus quantity (Delta) which can be calculated by using physical quantities of a lens. It is recognized that large objects in out-of-focused images can be restored well by using calculated PSF or OTF when a diaphragm is circular and (Delta) is estimated adequately. However details can not be restored well especially when (Delta) is large. Difference is recognized between PSF based on the geometrical optics and PSF based on the wave optics. It is recognized that images obtained by using PSF calculated by the wave optics are better in sharpness than by using PSF calculated by the geometrical optics. Such difference becomes larger as (Delta) becomes smaller. In the case that the diaphragm differs from a circular disc such as home video cameras, it is important to take the diaphragm form into account.

In this work we investigated the forensic applicability of some state-of-the-art image restoration techniques for digitized video-images and photographs: classical Wiener filtering, constrained maximum entropy, and some variants of constrained minimum total variation. Basic concepts and experimental results are discussed. Because all methods appeared to produce different results, a discussion is given of which method is the most suitable, depending on the image objects that are questioned, prior knowledge and type of blur and noise. Constrained minimum total variation methods produced the best results for test images with simulated noise and blur. In cases where images are the most substantial part of the evidence, constrained maximum entropy might be more suitable, because its theoretical basis predicts a restoration result that shows the most likely pixel values, given all the prior knowledge used during restoration.

A large number of algorithms based on partial differential equations (PDE) have recently been proposed to tackle the problems of noise removal, image enhancement and image restoration in real images. Starting with a noisy original image, the algorithms remove noise and enhance the original image by iterating the image using various schemes that are controlled by mean curvature, min/max flow, etc. We first present a variational approach such that during image restoration, edges detected in the original image are being preserved, and then we compare in a second part, the mathematical foundation of this method with respect to some of the well known methods recently proposed in the literature within the class of PDE based algorithms. The performance of our approach will be carefully examined and compared to some of the most recent algorithms proposed in the literature within the class of PDE based algorithms. Experimental results on synthetic and real images will illustrate the capabilities of all the studied approaches.

Video is used as recording media in surveillance system and also more frequently by the Swedish Police Force. Methods for analyzing video using an image processing system have recently been introduced at the Swedish National Laboratory of Forensic Science, and new methods are in focus in a research project at Linkoping University, Image Coding Group. The accuracy of the result of those forensic investigations often depends on the quality of the video recordings, and one of the major problems when analyzing videos from crime scenes is the poor quality of the recordings. Enhancing poor image quality might add manipulative or subjective effects and does not seem to be the right way of getting reliable analysis results. The surveillance system in use today is mainly based on video techniques, VHS or S-VHS, and the weakest link is the video cassette recorder, (VCR). Multiplexers for selecting one of many camera outputs for recording is another problem as it often filters the video signal, and recording is limited to only one of the available cameras connected to the VCR. A way to get around the problem of poor recording is to simultaneously record all camera outputs digitally. It is also very important to build such a system bearing in mind that image processing analysis methods becomes more important as a complement to the human eye. Using one or more cameras gives a large amount of data, and the need for data compression is more than obvious. Crime scenes often involve persons or moving objects, and the available coding techniques are more or less useful. Our goal is to propose a possible system, being the best compromise with respect to what needs to be recorded, movements in the recorded scene, loss of information and resolution etc., to secure the efficient recording of the crime and enable forensic analysis. The preventative effective of having a well functioning surveillance system and well established image analysis methods is not to be neglected. Aspects of this next generation of digital surveillance systems are discussed in this paper.

New image processing techniques may have significant benefits to law enforcement officials but need to be legally admissible in court. Courts have different tests for determining the admissibility of new scientific procedures, requiring their reliability to be established by expert testimony. The first test developed was whether there has been general acceptance of the new procedure within the scientific community. In 1993 the U.S. Supreme Court loosened the requirements for admissibility of new scientific techniques, although the California Supreme Court later retained the general acceptance test. What the proper standard is for admission of such evidence is important to both the technical community and to the legal community because of the conflict between benefits of rapidly developing technology, and the dangers of 'junk science.' The Reginald Denny beating case from the 1992 Los Angeles riots proved the value of computerized image processing in identifying persons committing crimes on videotape. The segmentation process was used to establish the presence of a tattoo on one defendant, which was key in his identification. Following the defendant's conviction, the California Court of Appeal approved the use of the evidence involving the segmentation process. This published opinion may be cited as legal precedent.

Adapted wave form analysis, refers to a collection of FFT like adapted transform algorithms. Given a signal these methods provide special matched collections of templates (orthonormal bases) enabling an efficient extraction of structural components. As a result various operations such as denoising, undesirable background suppression, sharpening and enhancement can be achieved efficiently. Perhaps the closest well known example of such coding method is provided by musical notation, where each segment of music is represented by a musical score made up of notes (templates) characterized by their duration, pitch, location and amplitude, our method corresponds to transcribing the music in as few notes as possible. Since noise and static are difficult to describe efficiently we obtain as a byproduct a denoised version of the sound. This transcription in a score can be developed into a mathematical musical orchestration as described below. The extension to images and video is straightforward we describe the image by collections of oscillatory patterns (paint brush strokes) of various sizes, locations, and amplitudes using a variety of orthogonal bases.

An intriguing question in the matching of objects with patterned injures in two and three dimensions is that of an appropriate metric for closeness -- is it possible to objectively measure how well an object 'fits' a patterned injury. Many investigators have suggested an energy-based metric, and have used such metrics to analyze craniofacial growth and anatomic variation. A strict dependence on homology is the primary disadvantage of this energy functional for generalized biological structures; many shapes do not have obvious landmarks. Some tentative solutions to the problem of landmark dependency for patterned injury analysis are presented. One intriguing approach comes from recent work in axiomatic vision. This approach has resulted in the development of a multiresolution medial axis for the extraction of shape primitives which can be used as the basis for registration. A scale-based description of this process can be captured in structures called cores, which can describe object shape and position in a highly compact manner. Cores may provide a scale- and shape-based method of determining correspondences necessary for determining the number and position of landmarks for some patterned injuries. Each of the approaches described are generalizable to higher dimensions, and can thus be used to analyze both two- and three- dimensional data. Together, they may represent a reasonable way of measuring shape distance for the purpose of matching objects and wounds, and can be combined with texture measures for a complete description.

In a short period of two weeks experiments had to be done for court. The order was to investigate the effects of ballpoints shot transorbitally by a crossbow. The use of a high speed video camera turned out to be valuable for detailed observation of the ballpoint during launching and penetration of a gelatine model and demonstration of the results in court.

Sensors are needed for concealed weapon detection which perform better with regard to weapon classification, identification, probability of detection and false alarm rate than the magnetic sensors commonly used in airports. We have concluded that no single sensor will meet the requirements for a reliable concealed weapon detector and thus that sensor fusion is required to optimize detection probability and false alarm rate by combining sensor outputs in a synergistic fashion. This paper describes microwave, millimeter wave, far infrared, infrared, x-ray, acoustic, and magnetic sensors which have some promise in the field of concealed weapon detection. The strengths and weaknesses of these devices are discussed, and examples of the outputs of most of them are given. Various approaches to fusion of these sensors are also described, from simple cuing of one sensor by another to improvement of image quality by using multiple systems. It is further concluded that none of the sensors described herein will ever replace entirely the airport metal detector, but that many of them meet needs imposed by applications requiring a higher detection probability and lower false alarm rate.

This paper discusses some new enabling technologies for law enforcement and security. A wide variety of concealed weapon detection systems are being investigated to determine the potential payoffs of employing these sensors to detect weapons concealed under a person's clothing. The enabling sensing mechanisms being studied include infrared, millimeter wave, acoustic, and x-ray sensors. The primary emphasis of this paper is on infrared and millimeter wave. A new technique for processing sensor data by segmenting and partitioning non-homogeneous images into homogeneous regions for later detection and identification processing is described. The name of this method is automated statistical characterization and partitioning of environments (A'SCAPE). A'SCAPE enables image enhancement for reliable detection and identification of weapons concealed under varying layers of clothing through its mapping process. By employing a variety of sensors, another enabling technology for concealed weapon detection (CWD) is sensor fusion. Concepts for experiments and analysis are discussed to determine the feasibility of various sensor fusion approaches for CWD.

The FBI's DRUGFIRE^TM system is a nationwide computerized networked image database of ballistic forensic evidence. This evidence includes images of cartridge cases and bullets obtained from both crime scenes and controlled test firings of seized weapons. Currently, the system is installed in over 80 forensic labs across the country and has enjoyed a high degree of success. In this paper, we discuss some of the issues and methods associated with providing a front-end semi-automated image capture system that simultaneously satisfies the often conflicting criteria of the many human examiners visual perception versus the criteria associated with optimizing autonomous digital image correlation. Specifically, we detail the proposed processing chain of an intelligent image capture system (IICS), involving a real- time capture 'assistant,' which assesses the quality of the image under test utilizing a custom designed neural network.

An x-ray image intensified cooled CCD camera (x-ray IICCCDC) with micro-focused x-ray source, is a very powerful tool for objects investigation. We can get inside images of objects non destructively and we can also get magnified images of the objects. We investigate illegal electronic circuits, forged IC chips, etc. by the x-ray IICCCDC. The data depth of each pixel of the image taken by the x-ray IICCCDC is 14 bits. The x-ray IICCCDC is a powerful tool for itself, but the raw images of that are degraded by many factors, a geometrical distortion, lack of uniformity of an x-ray window's material, blurring caused by finite size of the x- ray source, background noise by x-ray scattering, etc. Some of these degradation factors are well defined and the degradation can be restored well, but some of these degradation can not be recovered well. We analyzed the sources of degradation and applied digital image processing techniques to the images of the x-ray IICCCDC to enhance the image quality. Using these digital image processing techniques, the image quality of the x-ray IICCCDC was enhanced substantially.

In this paper, an original approach to deal with the important problem of stereovision using a weakly calibrated pair of images is presented. Given two different views of some 3D objects, and a virtual 3D plan set by an external operator, the method we propose allows to recover the 2D projections, in the two images, of the 3D planar curves corresponding to the intersection of the virtual plan with the different objects in the scene. To this end, an arbitrary curve is first initialized in one of the two images. This curve, and its associated homographic curve in the second image are then designed to move under the influence of internal and external image dependent forces while minimizing an energy functional. Following the work on geodesic active contours by Caselles et al and Malladi et al, we then transform the problem of minimizing this functional into a problem of geodesic computation in a Riemannian space, according to a new metric. The Euler- Lagrange equation of this new functional is derived and its associated PDE is then solved using the level set formulation scheme of Osher and Sethian by viewing it as a front propagating with internal and external image correlation dependent speed. The curves to be matched are therefore modelized as geodesic active contours evolving toward the minimum of the designed functional. Using this level set based approach, complex curves can be matched and topological changes for the evolving curves are naturally managed. The final result is also relatively independent of the curve initialization. Promising experimental results have been obtained on real images and some of these results are illustrated in the experimental section of this paper.

The anthropometry and movements are unique for every individual human being. We identify persons we know by recognizing the way the look and move. By quantifying these measures and using image processing methods this method can serve as a tool in the work of the police as a complement to the ability of the human eye. The idea is to use virtual 3-D parameterized models of the human body to measure the anthropometry and movements of a crime suspect. The Swedish National Laboratory of Forensic Science in cooperation with SAAB Military Aircraft have developed methods for measuring the lengths of persons from video sequences. However, there is so much unused information in a digital image sequence from a crime scene. The main approach for this paper is to give an overview of the current research project at Linkoping University, Image Coding Group where methods to measure anthropometrical data and movements by using virtual 3-D parameterized models of the person in the crime scene are being developed. The length of an individual might vary up to plus or minus 10 cm depending on whether the person is in upright position or not. When measuring during the best available conditions, the length still varies within plus or minus 1 cm. Using a full 3-D model provides a rich set of anthropometric measures describing the person in the crime scene. Once having obtained such a model the movements can be quantified as well. The results depend strongly on the accuracy of the 3-D model and the strategy of having such an accurate 3-D model is to make one estimate per image frame by using 3-D scene reconstruction, and an averaged 3-D model as the final result from which the anthropometry and movements are calculated.

A system is proposed that simultaneously captures the three- dimensional shape of a face and its surface texture. Such a three-dimensional model allows us to compare surveillance images of an offender irrespective of the pose of the offender's head. Also a single model for face albedo has been elaborated and its use will be demonstrated for viewing under different lighting conditions. The 3D acquisition system is based on an active technique, i.e. special illumination, but in contrast to traditional active sensing does not require scanning or sequential projection of multiple patterns. As a consequence, 3D photographs can be taken from a single image, and thus also when suspects do not collaborate.

Wavelet transform based techniques were developed and investigated for isolation and enhancement of objects in images. The primary motivation is the development of image processing algorithms as part of an automatic system for the detection of concealed weapons under a person's clothing; a problem of considerable potential utility to the military in certain common types of deployment in the post cold war environment such as small unit operations. The issue has potential for other dual use purposes such as law enforcement applications. Wavelet decompositions of the currently available images in the Rome Laboratory database, namely, noisy, low contrast, infrared images, were studied in space-scale-amplitude space. An isolation technique for separating potential suspicious regions/objects from surrounding clutter has been proposed. Based on the images available, the study indicates that the technique is promising in providing the image enhancement necessary for further pattern detection and classification.

This paper presents an approach to image fusion for concealed weapon detection (CWD) applications. In this work, we use image fusion to combine complementary image information from different sensors to obtain a single composite image with more detailed and complete information content. As a result of this processing, the new images are more useful for human perception and automatic computer analysis, tasks such as feature extraction and object recognition. In the fusion process, the images are first decomposed based on wavelet transform. Then at each lower resolution the images are fused by using several feature selection algorithms. The final composite image is obtained by taking the inverse wavelet transform of the fused wavelet coefficients. This technique has been applied to real data obtained from IR sensors. Special emphasis is placed on situations when weapons may not be completely visible from the sensors. Fusion results that demonstrate the utility of our approach are presented.