Image Technology

Mobile robots operate in a wide variety of environments, and the tasks they are being designed to perform vary from the simplest pick-and-place factory jobs to space station construction, maintenance, and repair. Vision systems for mobile robots are used to help locate goal objects or locations, plan paths to the goal, avoid obstacles along the chosen path, monitor the robot’s progress along the path, locate landmarks, recognize objects, compute motion parameters, etc. The vision capability required for any robot depends heavily on its environment and its assigned tasks. A robot that works in a factory picking and placing objects whose geometry, location, and orientation are known and constant may not need any vision capabilities at all. On the other hand, a robot working autonomously in a dynamic 3D environment like space or under water, with possibly unknown objects moving with arbitrary accelerations and rotations, in the presence of humans would need complex visual perception capabilities.

Deriving three-dimensional information about a scene from its images is a challenging problem in computer vision. Multiple views of a scene/object from a moving camera can be used for this task. This paper presents a critical appraisal of the theory and algorithms for problems involving 3D geometric constraints between a scene and its multiple views. The essential constraints between the 3D shape, the view transformations and the 2D image projections are presented for various widely applicable models of projection. The recent trend in representing 3D shape in a fixed object-centered coordinate system figures prominently in this paper. It is shown how this approach nicely separates the contribution of 3D shape and motion as manifested in the image motion. This is contrasted with the traditional approach of computing 3D motion as a precursor to the computation of 3D structure. Methods that incorporate these constraints using discrete features (like points) and spatio-temporal intensity gradients are presented.

Studies on visual perception have demonstrated that selective attention mechanisms and space-variant sensing are powerful tools for focusing available computing resources to the process of relevant data. In this paper an overall architecture for an active, anthropomorphic robot vision system which integrates retina-like sensing and attention mechanisms is proposed. Gaze direction is shifted both on the basis of sensory and semantic characteristics of the visual input, which are extracted separately by means of a parallel and serial analysis. An implementation of the system by means of optical flow and neural network techniques is described, and the results of its application are discussed.

This chapter reviews and evaluates color image models that have been used in machine vision. Color image formation is described using models for image sensors, surfaces, and reflection processes. These models have been used to predict properties of color pixel distributions that will result for various classes of scenes. Several algorithms are described that use these distribution models for applications such as image segmentation, illuminant color estimation, and illumination invariant recognition. For textured images that are common in outdoor applications, benefits can be derived from using spatial interaction models for color images. A detailed summary is presented of recent work that introduces color texture models and applies these models to image segmentation and geometry invariant surface identification.

We consider the problem of image segmentation and describe an algorithm that is based on the Minimum Description Length (MDL) principle, is fast, is applicable to multiband images, and guarantees closed regions. We construct an objective function that, when minimized, yields a partitioning of the image into regions where the pixel values in each band of each region are described by a polynomial surface plus noise. The polynomial orders and their coefficients are determined by the algorithm. The minimization is difficult because (1) it involves a search over a very large space and (2) there is extensive computation required at each stage of the search. To address the first of these problems we use a region-merging minimization algorithm. To address the second we use an incremental polynomial regression that uses computations from the previous stage to compute results in the current stage, resulting in a significant speed up over the non-incremental technique. The segmentation result obtained is suboptimal in general but of high quality. Results on real images are shown.

Counting the number of people crossing a public area can be very useful for properly scheduling the frequency of a service. Mechanical and photosensitive systems, such as rotating tripod gates, short iron doors, weight-sensitive boards, and photoelectric cells, have often been used for such estimates. Since these methods are not efficient in critical conditions, vision-based approaches have been provided. Many of them identify moving objects through a segmentation process. Once the objects are identified, they are tracked in the sequence of images and counted. These approaches have some drawbacks when they are used in critical conditions such as for counting the people getting on and off a public bus. In this paper, a new technique for counting passing people which is based on motion estimation and spatio-temporal interpretation of the estimated motion is proposed, with its implementation on prototype DSP-based architecture.

A system for crowding evaluation in complex environments is presented. The system acquires and processes data from a set of cameras monitoring an underground scene. The processing structure is modelled as a hierarchical Bayesian network of interacting nodes; each node aims at obtaining the probabilistic value of the number of people, detected within either local areas or the whole station, starting from suitable features extracted from images. Piece-wise linear models allow mapping from the feature value space to the number of people to be performed. The modelling algorithm, based on the Bellman Principle, is discussed. Results obtained after an extended test phase in a station of Genova’s underground are reported.

In this paper a system for the recognition of car license plates is presented. The aim of the system is to read automatically the Italian license number of a car passing through a toll-gate. A TV camera and a frame grabber card are used to acquire a rear-view image of the vehicle. First, a segmentation phase locates the license plate within the image. Then a procedure based on feature projection estimates some image parameters needed to normalize, by a bilinear resampling, the license plate characters. Finally, the character recognition extracts the desired information. Feature points and template matching operators are used to get a robust solution under multiple acquisition condition. The system is able to reject an image if some conditions are not met, as, for instance, the passage of a foreign car through the tollgate. A test has been done on more than three thousand real images acquired under different weather and illumination conditions. Within a test set the system has discarded all the unrecognizable images, i.e., not provided with the required characteristics. The recognition rate has been close to 91%. The final percentage may increase if a more accurate drawing of the templates is performed.

In this paper we present a control system by vision of hybrid circuit production. It concerns controlling the position of the circuit links and the quality of the solder joints. The control procedure consists of extracting the parameters characterising each type of circuit and comparing each circuit in relation to that of the prior known model. This system is integrated into the production line taking into account the constrained real time. These circuits are intended for industrial purposes, for professional equipment and the general public where the controls imposed are severe. This system therefore allows automatic control of electronic circuits performed manually up until this day.

A method has been developed to find the so-called missing dots in a heliotest strip. Heliotest is a test print method which is used to determine the quality of gravure print paper. The developed method is based on image processing and supervised learning, requiring a 386-based MS/DOS computer, a commercial image processing board and software, a commercial co-ordinate table, a CCD camera, and lighting equipment.

A vision system for our PISCIS project is described. This is a proposed project for the use of an untethered autonomous underwater vehicle (AUV) for pipeline inspection. The vision system is termed PVS (the PISCIS vision system). It will assist the AUV in finding and following pipelines. A salient feature of the PVS is that it is designed to find all pipelines within the field of view and, thus, the AUV can follow any of them. Robust image interpretation is important as humans can not interact with the PVS and correct errors. Furthermore, the image quality is reduced by backscatter, light absorption, a non-uniform background (the sea-bed), and marine material on the pipelines. The PVS is fixed to a vehicle and must rely on a heading sensor and an altitude sonar to match image features with pipeline models. The models are retrieved from a map based on the vehicle’s position.

Problems of texture analysis in industry are considered. First, a literature survey of proposed industrial applications is presented and, then, some popular texture measures which have been successfully used in various applications and new promising approaches proposed recently are described. Finally, a comparative study of the texture measures is carried out by using a classification principle based on comparing sample distribution of feature values to predefined model distributions with known true class labels.

Illumination-invariant image processing is an extension of the classical technique of homomorphic filtering using a logarithmic point transformation. In this paper, traditional approaches to illumination-invariant processing are briefly reviewed and then extended using newer image processing techniques. Relevant hardware considerations are also discussed including the number of bits per pixel required for digitization, minimizing the dynamic range of the data for image processing, and camera requirements. Three applications using illumination-invariant processing techniques are also provided.

This paper compares the suitability and efficacy of five algorithms for determining the peak position of a line or light stripe to subpixel accuracy. The algorithms are compared in terms of accuracy, robustness and computational speed. In addition to empirical testing, a theoretical comparison is also presented to provide a framework for analysis of the empirical results.

This paper presents a detailed summary of the properties and basic facts about spline spaces and their B-spline bases. Examination is made of the many different joint-continuity conditions. Geometric continuity constraints are of special interest, as they appear to satisfy visual needs of the human observers. Least-squares approximation of analytic curves and discrete point sets is also discussed. Special attention is devoted to the problem of selecting the bast fit to a closed curve, considering all the possible shifted parametric descriptions of the curve.

The measurement of surface topography is an important inspection task as it provides useful information for process and quality control. A candidate technique for such an application is confocal imaging. The advantages of confocal imaging are that it is a non-contact measurement, can be operated at high speed (greater than 10 megapixels/sec) and submicron resolution, and provides height information in multi-layered semi-transparent materials.

In this paper we present a system designed for fast acquisition and processing of confocal images. The system consists of an optical front end that uses tilted confocal scanning, and an image processing module. The function of the image processing module is to improve signal resolution, perform smoothing and detect surfaces in the incoming signal. The input signal is first deconvolved in order to improve the depth resolution, and then processed to identify significant peaks. These peaks represent the positions of different surfaces in the object being inspected. These peak locations are smoothed using a cluster based smoothing scheme to combat noise. For semi-transparent materials, our system is capable of detecting up to two surfaces at a given location.

Recognition of aircraft in complex, perspective aerial imagery is difficult because of occlusion, shadow, cloud cover, haze, seasonal variations, clutter and various forms of image degradation. This chapter describes a system for aircraft recognition that addresses some of these issues. The recognition system uses a hierarchical object model database that includes models represented using advance concepts to geometric entities. It involves three key processes: (a) The qualitative object recognition process is responsible for model-based symbolic feature extraction and generic object recognition; (b) The refocused matching and evaluation process accesses deeper levels of the database hierarchy with input from (a) to refine the extracted features and to perform more specific classification; and (c) the primitive feature extraction process regulates the extracted features based on their saliency and interacts with (a) and (b). Experimental results showing the qualitative recognition of aircraft in perspective, aerial images are presented.

Scanning probe microscopy (SPM) includes techniques such as scanning tunneling microscopy (STM), atomic force microscopy (AFM), magnetic force microscopy (MFM) and scanning ion conductance microscopy (SICM). Scanning probe microscopes have started a new era in microscopy by providing depth maps at an unprecedented resolution. These versatile devices work in vacuum, air, liquids, and aqueous solutions. Their resolution can be varied from the atomic range to the micrometer range. Scanning probe microscopy is being recognized as a powerful imaging technique in a variety of application areas. Not only can SPM image surface topography, but also other surface characteristics such as magnetic domains, electrical charge, local density of electron states, and surface temperature. Promising results using SPM have been obtained in imaging semiconductors, metals, organic materials, superconductors, and biological samples. SPM is already being used in some industrial applications and there is immense potential for applying it to surface characterization, metrology, and inspection in numerous applications. Image processing techniques are a vital complement to sensor technology in scanning probe microscopes. Image analysis and understanding techniques are essential if the potential of SPM for metrology and industrial inspection is to be realized. In this chapter, we present an overview of the state of the art in SPM with emphasis on image processing techniques for SPM. We outline the principle of operation of different scanning probe microscopes. Issues related to sensor technology are discussed. Commercially available scanning probe microscopes are listed and their features summarized. We review in detail the image processing work that has been done to date in relation to SPM and raise relevant issues. Existing and potential applications of SPM are discussed. Finally, we point out directions for future research in image processing related to SPM.

With recent advances in computer technologies, numerous new application areas requiring management of non-alphanumeric data such as images, videos, graphs, and audios have evolved. Examples of such applications include weather information management, medical information management, environmental pollution information systems, space exploration, manufacturing information management, genome research, training and educational systems, entertainment applications, and defense applications.

An efficient scheme based on an enhanced Laplacian pyramid (LP) is proposed for lossless/lossy compression and progressive transmission of medical images. The entropy of the LP is reduced by adopting two filters different for reduction and expansion. Encoding priority is given to major details through a hierarchical content-driven decision rule defining a binary quad-tree of split nodes, which is run-length encoded. The root layer of the pyramid is optimally chosen for spatial DPCM encoding. Error feedback along the layers of the LP ensures lossless and semi-lossy reconstruction capability and improves the robustness of the overall scheme. Reversible compression of scanned RX images, achieved at ratios of about 6:1, establish improvements over both DPCM and pyramid schemes. High-quality lossy versions at 60:1 compression ratio outperform JPEG both visually and quantitatively. Also results of NMR images are presented and discussed.

Broadly speaking, data compression strategies may be classified as either lossless or lossy. As its name suggests, the primary goal of lossless compression is to minimize the number of bits required to represent the source data without any loss of information. For image, and particularly video sources, however, some loss of information can usually be tolerated. There are three reasons for this. Firstly, significant loss of information can often be tolerated by the human visual system, without interfering with perception of the image or video sequence. Secondly, in many cases the digital input to the compression algorithm is, itself, an imperfect representation of a real-world scene. Thirdly, lossless image or video compression is usually incapable of satisfying the high compression requirements of most storage and distribution applications. In this chapter we are concerned only with lossy compression.