A prototype liquid-crystal-over-silicon spatial light modulator (SLM) with 16 x 16 pixels has been fabricated and used as an adaptive filter in the Fourier plane of a coherent optical processor. Preliminary results showing images of simple objects filtered through the SLM are presented. For comparison, images of the same objects obtained using photographic transparencies to simulate the action of the SLM as a filter are also presented.
High resolution imaging in ground-based astronomy is limited by the wave front degradation through the turbulent atmosphere. Adaptive optics offers the possibility to overcome these limitations. It employs a real-time phase compensation with a phase shifting optical element, which is usually a deformable mirror. The information for the control of the mirror surface is gained from a wavefront sensor. Thispaper describes an adaptive optics prototype for infrared imaging using a visible wavefront sensor and its expected performances in astronomy.
Present Artificial Intelligence development and its technical means are analyzed critically from the viewpoint of modelling human cognitive processes. It is shown, that the dominant present expert systems paradigm cannot prove successful in solving Artificial Intelligence global problems, unless it implement certain isomorphism with higher brain functions, suggested by brain science. A new firmware architecture alternative is suggested, integrating traditional information processing approaches with holographic ideology.
A VME-bus image pipeline processor for extracting vectorized contours from grey-level images in real-time is presented. This 3 Giga operation per second processor uses large kernel convolvers and new non-linear neighbourhood processing algorithms to compute true 1-pixel wide and noise-free contours without thresholding even from grey-level images with quite varying edge sharpness. The local edge orientation is used as an additional cue to compute a list of vectors describing the closed and open contours in real-time and to dump a CAD-like symbolic image description into a symbol memory at pixel clock rate.
The continuing miniaturisation of electronic components is pushing back the frontiers of real-time digital image processing so far as improving image quality, image coding, quantitative analysis in microscopy, analysis of X-ray images and multispectral remote sensing of the Earth are concerned. In all cases the image processing is heavily influenced by the sensing technique, and this paper aims to present the special features of IR image processing that have a bearing on this.
During the past few years new techniques based on circular harmonic expansion have been introduced in the field of pattern recognition and some important advances have been achie-ved recently. Whether using single circular harmonic filters or using multimode filters it is necessary to choose a geometrical center for the harmonic expansion. We show that this choice results in different performances when the designed filter is used, both with respect to the height of the correlation peak and also with respect to the discrimination capability between simi-lar inputs. For the single harmonic filter, we study the influence on the performance of the method of the following factors: the choice of the harmonic order and the choice of expansion cen-ter used. For multimode filters, a possible real time method for rotationally invariant pattern recognition can be developed on the basis of the detection of N correlation plane intensity matrix at the same time the filter spins a whole turn. In this case, the dependence on the choice of expansion center and the minimum number N of correlations for a good detection is shown.
The EDI system, operating at the Casaccia Centre (Rome), was set up assembling specific S/W and H/W facilities in order to obtain a system for:
- the acquisition of digital images (DIs),
- the processing of DIs using appropriate algorithms according to user's need,
- the releasing of the results (new images, graphs or numerical entities) on the most suitable supports.
This paper reports the performance evaluation of a matched filter for IR background suppression and target detection in one atmospheric window of the Middle Infrared Region. This filter is for application in an airborne staring infrared missile warning (IRMW) system. It is aimed at the possibility of separating the temporal fluctuation of background noise, induced by aircraft motion, from a particular short-time characteristic of the infrared missile threat signature. At first, mathematical models were applied to determine the main parameters and relations that affect background noise in relation to the system's relative motion in a selected scenario, as well as threat signature characteristics. In this phase of the study, a general expression for the optimum matched filter transfer function was obtained with an associated sub-optimal solution. Performances of these filters in many cases of interest are also evaluated and compared with those of a double-difference filter. Finally, experimental measurements were carried out with a scanning radiometer of IR background radiance, which provided the input for computer simulations to validate the mathematical models utilized.
A grey-tone image can be viewed as a three-dimensional continuous surface with peaks, ridges, valleys, plains and oriented facets. In this paper, simple methods are described to analyze the structure of this three-dimensional surface. A small number of classes are defined to represent the structure of the image, and an algorithm is used to classify each pixel according to its local grey-tone structure. Then, region growing procedures are given to segment the image into semantically meaningful regions. The problem of establishing a relationship between a topographic class and the physical properties of the image is also addressed and an algorithm is proposed to extract image patches which correspond to three-dimensional continuous surfaces in the scene. Applications of these algorithms to several images are presented and discussed.
A code was developed for generating X-ray images of bodies of any shape and spatial orientation. The bodies may be concave and have cavities. The geometrie of the bodies is defined by wire models which are displayed on the screen (in central projection) and can be modified by standard operations. The code allows for the following influencing effects:
-- wavelength and wavelength distribution of X-rays
-- quantity of absorbed photons
-- wavelength dependent absorption coefficient of the material -- characteristics of the 2-dimensional detector array (e.g. dynamic range,
-- quantum resolution, and noise)
By taking into account the kinematics of the body (translation and rotation) image sequences showing realistic features can be generated. Therefore, the code is a most useful tool for testing of pattern recognition and tracking algorithms, particulary when using 3-D-kinematic predictor procedures. The method is applicable e.g. in ballistics or to non-destructive material testing of fast moving machine parts. The application of this code is not restricted to radiography. It may also be used in topics like absorption of infrared radiation in aerosols, light density variationes in fluid flows and similar phenomena.
In many image processing, the image function values located between sucessive points are needed to be know with a sufficient precision. Indeed, in many applications such as robotics, biomedical imagery or process control, images need to be transformed before being processed. Such transformations allow to make original images closer to reference images. In this paper we propose some image transformation algorithms using spline function interpolation. For the main point, their application domains are image restoring, geometrical transformations or spatial distorsion processing. Comparisons are made among different spline functions to determine which of them is the best solution considering the computation complexity and the processing method efficiency. We also propose an edge detection method using this kind of functions.
The task of document recognition requires the scanning of a paper document and the analysis of its content and structure. The resulting electronic representation has to capture the content as well as the logic and layout structure of the document. The first step in the recognition process is scanning, filtering and binarization of the paper document. Based on the preprocessing results we delineate key areas like address or signature for a letter, or the abstract for a report. This segmentation procedure uses a specific document layout model. The validity of this segmentation can be verified in a second step by using the results of more time-consuming procedures like text/graphic classification, optical character recognition (OCR) and the comparison with more elaborate models for specific document parts. Thus our concept of model driven segmentation allows quick focussing of the analysis on important regions. The segmentation is able to operate directly on the raster image of a document without necessarily requiring CPU-intensive preprocessing steps for the whole document. A test version for the analysis of simple business letters has been implemented.
The paper describes an algorithm for recognition of 2-D objects which is insensible to their size and orientation. The criterion of similarity is the normalized dispersion of ratios of unknown and known object specific vector components. Components of characteristic vectors are represented by a mesh of vector lengths of fixed angle from objects center to their boundary with object orientation as the reference vector direction.
After a short introduction including a historical review of our experience up to now, the report describes some of the important possible applications of modern image processing in future helicopters. The tasks of the image processing system as applied to actual operational functions will be described for a wide range of applications, such as scene analysis on the field of battle, missile guidance for guided missiles such as HOT, multiple target tracking, alignment of ATGW 3 missiles with the gunner's sight and distance-specific algorithms for seeker heads and terminal-phase guidance. In the second part of the report, suitable and tested methods of realizing these applications will be described, with particular attention given to realization of a complex 20ms realtime algorithm enabling a target in an involved scenario, with several obstacles, to be successfully tracked. Lastly, we look at the algorithms required for seeker heads and the extreme miniaturization that their realization inevitably entails. The application examples described in this paper refer to work carried out by AEG or MBB, or in some cases to joint enterprises undertaken by both companies.
We propose the pyramidal BLI (Binary Laplacian Image) graph matching method for stereo vision, which uses the local as well as the global similarities to assure a good precision of matching results and to eliminate the ambiguities. Because the BLI is detected by DRF method which has a fast realization and matching between graphs is fast, a pseudo-real time system is possible.
A new approach for locating a mobile robot in a real indoor scene from a monocular image by natural frames is presented. Based on a fast vanishing points finding algorithm, the three rotational transform are determined. Meaningful and structured features which are amenable to high level interpretation are constructed, from linear features extracted from contours up to junction and face features. The high level interpretation about the height of face feature or junction feature permit to determine the translational transform.
The spatial coordinates of a known object relative to the camera position can be determi-ned from the analysis of a monocular view of an object. Prerequisite is that the objects in the scene can be identified. The objects, which may appear in the scene, are internally represented as a list of spatial form-elements. The presented method is a two-step procedure. After characteristic form-elements have been extracted from the image, the most plausible object position is calculated on the basis of a relaxation method. Different measures like the implementation of the relaxation process on parallel processors permit the determination of the spatial position in an industrial scene in real-time.
A method to determine a projection model from a sequence of TV images is presented. The projection model represents a plausible description of the changes wich occur in the image sequence. The algorithm can model moving rigid, opaque objects with a diffuse reflecting (matte) surface. The illumination of the scene must be diffuse. The algorithm finds a 3-D model which can be used to reconstruct a accurately 2-D projections of moving objects. Applications can be seen in the areas of image data compression, image sequence transmission with very low data rates and modelling of natural objects for Computer Graphics.
We describe a real-time optical device for vehicle detection and recognition on roads. With vehicle traffic on public roads steadily increasing, there is a growing need for efficient monitoring and control of traffic flow. There is need for traffic counting (vehicle detection), estimation of vehicle speed and for vehicle classification (i.e. passenger car, truck). An optical sensor is very well suited to this problem giving detailed information from vehicles seen from a position beside or above the road. We implemented a device for this task operating in real-time with image scanning rates of about 100 Hz. Such rates are necessary for vehicle speeds up to 100 km/h. We solved the problem of high data rates by reducing the sensor's field of view down to two parallel receptor columns with a well known spacing and by sampling the grayvalues from the columns with a rate well fitted to the minimum of the required information. The principal idea of the applied image processing is to use change detection by means of difference pictures, embedded in a dynamic scene driven control. For the reliable detection and segmentation of moving vehicles a simple vehicle model and some rough estimates of the traffic flow are used. The segmented image data of the vehicles together with the derived information are collected for further inspection and processing. The system is realized on the Visual Interpretation System for Technical Applications (VISTA) developed by the Fraunhofer Institute IITB /1/. Most data processing is performed by software. The system achieves well the picture scanning rate of 100 Hz. For experimental purposes all image processing data are continuously displayed on a video monitor.
We have studied the possibility of using microscopic interferometric techniques to monitor the growth of protein crystals on the Space Station. Nucleation in the solution is a critical phase and needs to be monitored carefully. It involves detecting and localizing nucleation sites in their earliest phase with dimensions of the order of 50-100nm. A comparison with the well-known technique of scattering and microscopic interferometry for monitoring the nucleation was made and the latter seems to be much better suited for this monitoring task. In microscopic interferometry, a detailed map of the solution density is made in the form of an interference pattern. From this pattern one should be able to detect the onset of nucleation. This technique offers several advantages, however, for space applications it is essential that the interferograms must be interpreted in an automated real-time manner. We have used techniques of digital image processing to develop a system for the real-time analysis of microscopic interferograms of nucleation sites during protein crystal growth. Some details of the optical setup and the image processing system along with experimental results will be presented.