This paper presents a new method for the extraction and clustering of Catadioptric Line Images (CLIs) from a single catadioptric image. The estimation of vanishing directions is also achieved simultaneously. With the mirror parameters provided, the central catadioptric system can be fully calibrated using the projected mirror boundary. Once the system is fully calibrated, the problem of conic fitting is reduced from five parameters to two parameters. This method requires the central catadioptric system to be vertically aligned i.e. the mirror axis is orthogonal to the floor. In a man-made environment where lots of parallel and orthogonal lines are presented, placing the catadioptric system in such a way allows us to obtain some degenerate CLIs, e.g. projections of lines orthogonal or parallel to the mirror axis. In this paper, we integrate the geometric properties associated with the degenerate CLIs into a J-linkage based algorithm for CLIs clustering and Vanishing Points (VPs) estimation. An exhaustive analysis on the influence of arc occlusion and noise to CLI fitting and grouping is also performed. The proposed approach is targeted to tackle the problem of occlusion and noise. The evaluation of the method is carried out on both synthetic and real data.

Building recognition is an important task for a wide range of computer vision applications, e.g., surveillance and intelligent navigation aid. However, it is also challenging since each building can be viewed from different angles or under different lighting conditions, for example, resulting in a large variability among building images. A number of building recognition systems have been proposed in recent years. However, most of them are based on a complex feature extraction process. In this paper, we present a new building recognition model based on local oriented features with an arbitrary orientation. Although the newly proposed model is very simple, it offers a modular, computationally efficient, and effective alternative to other building recognition techniques. According to a comparison of experimental results with the state-of-the-art building recognition systems, it is shown that the newly proposed SFBR model can obtain very satisfactory recognition accuracy despite its simplicity.

Building recognition is a nontrivial task in computer vision research which can be utilized in robot localization, mobile navigation, etc. However, existing building recognition systems usually encounter the following two problems: 1) extracted low level features cannot reveal the true semantic concepts; and 2) they usually involve high dimensional data which require heavy computational costs and memory. Relevance feedback (RF), widely applied in multimedia information retrieval, is able to bridge the gap between the low level visual features and high level concepts; while dimensionality reduction methods can mitigate the high-dimensional problem. In this paper, we propose a building recognition scheme which integrates the RF and subspace learning algorithms. Experimental results undertaken on our own building database show that the newly proposed scheme appreciably enhances the recognition accuracy.

The problem of efficient bit allocation for dependent video coding within the context of ascribing a quantizer to each macroblock (MB) within a frame is addressed in this letter. Strongly dependent MBs exhibit significant discontinuities on their operational distortion-rate (D-R) convex hulls that nullify the assumptions made for the use of the coordinate-wise steepest-descent (SD) algorithm. A qualified slope concept and a near-neighborhood look-ahead modification are introduced to circumvent these problems and are experimentally evaluated on a set of H.263+ encodings.

Maximizing the potential of latent fingerprints from crime scenes in the identification of suspects requires the rapid transfer of the latent from the scene to a remote fingerprint bureau. Transmission over restricted-bandwidth cellular wireless networks requires the latent images to be compressed but without compromising the likelihood of a match being achieved. We present details of experiments to establish the optimum form of compression that provides realistic transmission times and yet does not affect the utility and integrity of the U.K. Fingerprint Service in searching for latent identifications and in archiving unidentified latents on the U.K. national automatic fingerprint identification system (AFIS). Practical aspects of the implemented system, especially in respect to communication and security protocols, are outlined. Finally, we give some details of the operational advantages of this system as it begins to be employed across U.K. police forces.

Relevance feedback (RF) schemes based on support vector machines (SVMs) have been widely used in content-based image retrieval (CBIR). However, the performance of SVM-based RF approaches is often poor when the number of labeled feedback samples is small. This is mainly due to 1) the SVM classifier being unstable for small-size training sets because its optimal hyper plane is too sensitive to the training examples; and 2) the kernel method being ineffective because the feature dimension is much greater than the size of the training samples. In this paper, we develop a new machine learning technique, multitraining SVM (MTSVM), which combines the merits of the cotraining technique and a random sampling method in the feature space. Based on the proposed MTSVM algorithm, the above two problems can be mitigated. Experiments are carried out on a large image set of some 20,000 images, and the preliminary results demonstrate that the developed method consistently improves the performance over conventional SVM-based RFs in terms of precision and standard deviation, which are used to evaluate the effectiveness and robustness of a RF algorithm, respectively.

A self-organizing mixture network (SOMN) is derived for learning arbitrary density functions. The network minimizes the Kullback-Leibler information metric by means of stochastic approximation methods. The density functions are modeled as mixtures of parametric distributions. A mixture needs not to be homogenous, i.e., it can have different density profiles. The first layer of the network is similar to Kohonen's self-organizing map (SOM), but with the parameters of the component densities as the learning weights. The winning mechanism is based on maximum posterior probability, and updating of the weights is limited to a small neighborhood around the winner. The second layer accumulates the responses of these local nodes, weighted by the learned mixing parameters. The network possesses a simple structure and computational form, yet yields fast and robust convergence. The network has a generalization ability due to the relative entropy criterion used. Applications to density profile estimation and pattern classification are presented. The SOMN can also provide an insight to the role of neighborhood function used in the SOM.

This article considers the dynamics of speculative trading ranges. Daily trading ranges provide good estimates of the level of speculative volatility, and analysis of the daily trading range of twenty US futures markets finds that first order differences of the logarithm of daily range show significant negative autocorrelation. This mean-reverting process is also revealed with Hurst analysis. Spectral analysis shows that the underlying dynamics of speculative trading ranges is a pink noise process with each futures market yielding a spectral exponent below that of brown noise.

Hyvarinen et al. (2000) have developed an ICA based method for image denoising. The major advantage of their method is that the transformation matrix can by adjusted to suit the available data. However, in their method, the transformation matrix and shrinkage parameters need to be learned from noise-free data. In this paper, we propose a simplified shrinkage scheme, which has only one heuristic control parameter. Experimental results show that the ICA based method with this new shrinkage scheme achieves comparable performance to that of Hyvarinen et al.

Increasingly the problems of recognising faces under a variety of viewing conditions, including depth rotations, is being considered in the field. The concept of norm-based coding in face recognition is not new but has been little investigated in machine models. Here we describe a norm-based face recognition system which is capable of generalising from a single training view to recognise novel views of target faces. The system is based upon the characteristic nature faces as they move through a pose-varying eigenspace of facial images and deviations from the norm of a gallery of face images. We illustrate the use of the technique for a large range of pose variation.

The self-organising mixture network (SOMN) is a learning algorithm for mixture densities, derived from minimising the Kullback-Leibler information by means of stochastic approximation methods. It has been shown the SOMN converges faster than the EM-based algorithms and generalises better as it is based on the expected likelihood rather than the sample likelihood. The derived algorithm has similar updating forms to the self-organising map (SOM), thus reveals the mixture interpreter role of the neighbourhood function used in the SOM. When the sample set is small, overfitting problems often occur in most algorithms. Further improvement can be achieved by averaging ensembles of the SOMNs. The algorithms have been applied to both experimental data and real-world problems. The results show that smoothed mixtures with improved accuracy have been obtained. Estimation variance has been reduced.

A working face recognition system requires the ability to represent facial images in such a way that permits efficient and accurate processing. The human visual system effectively stores, recognises and classifies familiar facial images under a wide variety of viewing conditions, albeit with various degrees of accuracy. We describe a system which automatically determines a representation for pose-varying facial images - a representation with inherent classification properties, an ability to generalise from one viewing condition to another, and which uses fast computational procedures. 1 Introduction Representing a human face under a large range of viewing conditions and being able to recognise images of familiar people in these conditions is one of the more remarkable abilities of the human cognitive system. Whilst the relatively simple procedure of matching frontal face images under consistent viewing conditions has been implemented and commercially exploited [17, 23], we have yet to...

In this paper an analysis of the statistical and the convergence properties of Kohonen's self-organizing map of any dimension is presented. Every feature in the map is considered as a sum of a number of random variables. We extend the Central Limit Theorem to a particular case, which is then applied to prove that the feature space during learning tends to multiple gaussian distributed stochastic processes, which will eventually converge in the mean-square sense to the probabilistic centers of input subsets to form a quantization mapping with a minimum mean squared distortion either globally or locally. The diminishing effect, as training progresses, of the initial states on the value of the feature map is also shown.

A combination of the self-organising principle and a relaxation method is proposed for the unsupervised segmentation of textured images. There are two phases to this segregation. The first one is region-based and provides a coarse rapid segmentation. A hierarchical self-organising learning structure together with a Markov random field (MRF) model-based LS estimator is used for parameter estimation and classification of different regions from a randomly located window, which shrinks with time. A globally correct separation of different regions has been always formed during extensive experiments. These results are used as the initial states for the second phase processing, which is boundary-based. A relaxation method, similar to the Metropolis algorithm, is used for boundary improvement. Over a small area of original boundary, a new boundary is randomly generated, which accepted or rejected according to the local energy change. Experiment results on both synthetic and real textured images are presented.>

The fine granularity and reconfigurable nature of field-programmable gate arrays (FPGA's) suggest that defect-tolerant methods can be readily applied to these devices in order to increase their maximum economic sizes, through increased yield. This paper identifies the inability to contain faults within single cells and the need for fast reconfiguration as the key obstacles to obtaining a significant increase in yield. Monte Carlo defect modeling of the photolithographic layers of VLSI FPGA's is used as a foundation for the yield modeling of various defect-tolerant architectures. Results suggest that a medium-grain architecture is the best solution, offering a substantial increase in size without significant side effects. This architecture is shown to produce greater gate densities than the alternative approach of realizing ultralarge scale FPGA's-multichip modules.>

We describe a technique for matching a single, learned elastic model of the shape of normal chromosomes to chromosomal images. Our model has a hierarchical organisation, with increasingly coarse shape descriptions at higher levels. A problem of finding the model description most likely to have generated an image is reduced to one of matching the locations of model points to the locations of image features encoded on a Kohonen Self-Organising Map (SOM). During matching, coarse shape information is communicated between levels via fixed, viewpoint-independent transformations between object-based frames. After matching, frame parameters provide a compact, multi-scale description of important shape information. We propose using this information to train a neural network to recognise structurally damaged chromosomes.