Showing papers in "Pattern Recognition Letters in 1992"

TL;DR: A new algorithm is presented for local maximum and minimum filters requiring only 6 comparisons per pixel independent of kernel size, based on separability and a combination of block recursive series which are evaluated forwards and backwards.

TL;DR: A faster nearest neighbor search method for image data vector quantization that uses hyperplanes orthogonal to the central line direction of the coordinate system to partition the search space.

TL;DR: The use of the Euclidean distance map guarantees obtaining a skeleton with structure sufficiently stable under object rotation, and placed where it is expected to be found according to human intuition.

TL;DR: The results suggest that the minimum adequate number of training samples is at least ten times the number of classifier parameters.

TL;DR: A histogram-based approach to thresholding, called hill-clustering, is developed, which is computationally fast and efficient and the variation of computational time of this proposed approach is insignificant regardless of the number of thresholds to be identified.

TL;DR: The present technique introduces the concept of an asymmetric region of support and k - l -cosine, which needs no input parameter and remains reliable even when features of multiple size are present.

TL;DR: A tailored version of the probability measure of a fuzzy event is used for image segmentation and its application to clustering problems is indicated and applied to an object extraction problem.

TL;DR: A new method, based on k -cosine, is proposed for the determination of region of support and detection of significant points of digitized curves and a polygonal approximation is suggested by joining the successive significant points.

TL;DR: The proposed algorithm performs ordered propagation using Euclidean distance transformation without generating any distance map, which allows optimization of both the time and memory demand.

TL;DR: It was found that different elliptical-parameter-estimation techniques must be applied depending on acceptable computational cost, number of parameters to be estimated, the required degree of accuracy, and the specific conditions under which the estimation must be performed.

TL;DR: A new parallel thinning algorithm for binary images employs template matching to remove the edge points of an object shape in a binary image iteratively and obtains skeletons with the properties of perfectly 8-connected, noise-insensitive, and topologically equivalent to the original object shape without excessive erosion.

TL;DR: The meaning of performance characterization in general is expanded, the experimental protocol under which an algorithm performance can be characterized is discussed, and the discussion to the area of thinning algorithms as a case in point is focused on.

TL;DR: In this paper, an efficient approximation elimination search algorithm for fast nearest-neighbour search is proposed based on a spherical distance coordinate formulation, where a vector in K -dimensional space is represented uniquely by its distances from K + 1 fixed points.

TL;DR: It has been found that distinguishing between different types of features in a model or scene results in a very efficient implementation of Geometric Hashing using a multidimensional hash table, and the filtering ratio of this scheme turns out to be high enough to allow raliable recognition with the corerct feature correspondence between model and scene.

TL;DR: A closed-form formula for the average number of quadtree nodes (‘pieces’ or ‘blocks’) required by a rectangle aligned with the axes is given, which can be generalized for k-dimension spaces (k = 1,3,…), allowing the study of segment trees, oct-trees etc.

TL;DR: A novel learning algorithm is proposed, called Back-Propagation for Sequences (BPS), for a particular class of dynamic neural networks in which some units have local feedback, and it has the same time complexity and space requirements as back-propagation (BP) applied to feedforward networks.

TL;DR: This paper develops a new method for intensity image segmentation using a neural network of non-linear analog neurons modelled as amplifiers which have sigmoid monotonic input-output relations.

TL;DR: A method based on an analysis of the shape of a word as a whole object demonstrated to be a useful alternative for recognizing degraded word images that are prone to errors in character segmentation.

TL;DR: Algorithms for object extraction using a neural network using a single neuron (processor) to every pixel for its operation in order to implement the concept of self-organized feature mapping are proposed.

TL;DR: A method for decomposing a digital image binary clump into constituent convex parts is described, based on a novel contour tracing and labelling technique which simultaneously erodes selective parts of the clump while checking for splitting into disjoint blobs.

TL;DR: Two distribution-free upper bounds are given for the true error rate of a classifier, using the resubstitution error estimate, improving a bound proposed by Vapnik (1982).

TL;DR: A new algorithm for Artificial Neural Nets which behaves well on complex data sets and is very fast compared with Back-Propagation— improvements of four orders of magnitude have been obtained.

TL;DR: An algorithm for efficient classification of data in R μ when there exists no a priori information about the number of clusters is developed, and a theorem on the convergence of this algorithm is proved.

TL;DR: A new method for polygonal approximation based on Pavlidis' concept of ‘almost collinearity’ of a set of points, applicable to digital curves with uniformly spaced points as well as non-uniformly spaced points.

TL;DR: A new method is proposed for detecting and locating objects of interest within a colour scene under very strong variabilities in lighting conditions, object shape and pigmentation using Nearest Neighbour classification and Multiedit-Condensing techniques.

TL;DR: A new technique for designing binary decision tree classifiers that is chosen at each node in such a way that the classifier formed by combining the decision trees behaves in an optimal fashion is described.

TL;DR: Some combinatorial properties of two planar Euclidean proximity graphs, the relatively closest graph and the modified Gabriel graph, are derived.

TL;DR: Using lattice-theoretic results, an efficient algorithm is given for computing the shortest path distance between two binary trees using a transformation on rooted unlabeled binary trees.

TL;DR: A new technique called ‘dynamical threshold’ is designed to extract the correct features of original patterns and provides a method to segment complex stroke structures into strokes from which the structures are composed.