International Journal of Pattern Recognition and Artificial Intelligence

We present a system for vision guided autonomous circumnavigation, allowing a mobile robot to navigate safely around objects of arbitrary pose, and avoid obstacles. The system performs model-based object recognition from an intensity image. By enabling robots to recognize and navigate with respect to particular objects, this system empowers robots to perform deterministic actions on specific objects, rather than general exploration and navigation as emphasized in much of the current literature. This paper describes a fully integrated system, and, in particular, introduces canonical-views. Further, we derive a direct algebraic method for finding object pose and position for the four-dimensional case of a ground-based robot with uncalibrated vertical movement of its camera. Vision for mobile robots can be treated as a very different problem to traditional computer vision, as mobile robots have a characteristic perspective, and there is a causal relation between robot actions and view changes. Canonical-views are a novel, active object representation designed specifically to take advantage of the constraints of the robot navigation problem to allow efficient recognition and navigation.

This paper further explores the uses of Bayesian Networks for detecting buildings from digital orthophotos. This work differs from current research in building detection in so far as it utilizes the ability of Bayesian Networks to provide probabilistic methods for evidence combination and, via training, to determine how such evidence should be weighted to maximize classification. In this vein, then, we have also utilized expert performance to not only configure the network values but also to adapt the feature extraction pre-processing units to fit human behavior as closely as possible. Results from digital orthophotos of the Washington DC area prove that such an approach is feasible, robust and worth further analysis.

The goal of this work was to develop an improved defect detection scheme for high-speed real-time web surface inspection. This goal was realized by splitting the task into two independent parts: feature extraction and segmentation. Both parts were implemented using efficient algorithms which were implemented in hardware that is suitable and fast enough to be included in a working web inspection system. The proposed scheme is based on some derived texture features and a new self-organizing map variant, the statistical self-organizing map. These techniques offer several improvements over the gray-level thresholding techniques that have been traditionally used in commercial web inspection systems.

Lines and edges play an important role in pattern recognition because they mark object surfaces and boundaries. In spite of many attempts to construct optimal detectors, e.g. an edge detector, it appears that all known algorithms have problems at locations where lines and edges are very close and/or intersect. Furthermore, there are still very few schemes which can detect and classify lines and edges (events) simultaneously.

In view of the fact that the human visual system does not seem to suffer from any problems, our aim is to develop an event detection scheme that makes use of biologically-motivated operators and, therefore, to overcome the problems known from the literature. In this scheme we apply 2D complex Gabor filters and exploit the local phase information. In order to cope with events having a different curvature, we develop an adaptive scheme that is based on compound Gabor filter kernels with different orientation bandwidths.

This paper describes a novel method for edge feature detection of document images based on wavelet decomposition and reconstruction. By applying the wavelet decomposition technique, a document image becomes a wavelet representation, i.e. the image is decomposed into a set of wavelet approximation coefficients and wavelet detail coefficients. Discarding wavelet approximation, the edge extraction is implemented by means of the wavelet reconstruction technique. In consideration of the mutual frequency, overlapping will occur between wavelet approximation and wavelet details, a multiresolution-edge extraction with respect to an iterative reconstruction procedure is developed to ameliorate the quality of the reconstructed edges in this case. A novel combination of this multiresolution-edge results in clear final edges of the document images. This multi-resolution reconstruction procedure follows a coarser-to-finer searching strategy. The edge feature extraction is accompanied by an energy distribution estimation from which the levels of wavelet decomposition are adaptively controlled. Compared with the scheme of wavelet transform, our method does not incur any redundant operation. Therefore, the computational time and the memory requirement are less than those in wavelet transform.

In recent years, wavelets have attracted great attention in both still image compression and video coding, and several novel wavelet-based image compression algorithms have been developed so far, one of which is Shapiro's embedded zerotree wavelet (EZW) image compression algorithm. However, there are still some deficiencies in this algorithm. In this paper, after the analysis of the deficiency in EZW, a new algorithm based on quantized coefficient partitioning using morphological operation is proposed. Instead of encoding the coefficients in each subband line-by-line, regions in which most of the quantized coefficients are significant are extracted by morphological dilation and encoded first. This is followed by using zerotrees to encode the remaining space which has mostly zeros. Experimental results show that the proposed algorithm is not only superior to the EZW, but also compares favorably with the most efficient wavelet-based image compression algorithms reported so far.

This paper presents a novel methodology of disambiguating prepositional phrase attachments. We create patterns of attachments by classifying a collection of prepositional relations derived from Treebank parses. As a by-product, the arguments of every prepositional relation are semantically disambiguated. Attachment decisions are generated as the result of a learning process, that builds upon some of the most popular current statistical and machine learning techniques.

We have tested this methodology on (1) Wall Street Journal articles, (2) textual definitions of concepts from a dictionary and (3) an ad hoc corpus of Web documents, used for conceptual indexing and information extraction.

We continue the study of pictorial languages as formalized in Ref. 1 (based on the operations of shifting and superposing elementary pictures). If the pixels are superposed without composing their colors, then we produce only recursive languages. When the colors of the superposed pixels can be composed, then any array grammar can be simulated (hence, all recursively enumerable languages can be obtained). Bidimensional pictorial frameworks with a nonrecursive membership problem are obtained in the restricted case when (1) we do not allow the superposition of nontransparent pixels, excepting the fact that (2) for each color there is a complementary color, which, superposed on the original one, leads to a transparent pixel.