Least desirable feature elimination in a general pattern recognition problem

A new technique for dimensionality reduction for a general pattern recognition problem with pattern classes represented by multivariate normal distributions is presented. The method consists of identifying and eliminating the least desirable feature out of the original feature space. It is shown that great simplicity is obtained by doing so in comparison to the existing methods. A very simple expression describing a vector representing the least desirable feature in a most general case is derived. If the original feature space is N-dimensional then recognizing and eliminating such a feature is equivalent to selecting and retaining the N−1 best features. J-divergence is used as a measure of the discrimination between the classes. The flow chart for the method for discarding more than one dimension is presented.     

Natural language understanding by a robot: A pattern recognition problem

We present a program that understands Natural Language Commands, i.e. which have a free form but a well-defined semantic content. We show that by Structural Pattern Recognition techniques, the program “learns” the vocabulary and the structure of the sentences. An example is given, “the taxi driver robot”, which is directed by Natural Language Commands.     

Modeling broken characters recognition as a set-partitioning problem

This paper presents a novel technique for recognizing broken characters found in degraded text documents by modeling it as a set-partitioning problem (SPP). The proposed technique searches for the optimal set-partition of the connected components by which each subset yields a reconstructed character. Given the non-linear nature of the objective function needed for optimal set-partitioning, we design an algorithm that we call Heuristic Incremental Integer Programming (HIIP). The algorithm employs integer programming (IP) with an incremental approach using heuristics to hasten the convergence. The objective function is formulated as probability functions that reflect common OCR measurements – pattern resemblance, sizing conformity and distance between connected components. We applied the HIIP technique to Thai and English degraded text documents and achieved accuracy rates over 90%. We also compared HIIP against three competing algorithms and achieved higher comparative accuracy in each case.     

Stochastic reasoning for structural pattern recognition: An example from image-based UAV navigation

This paper reports on the statistical embedding of a structural pattern recognition system into the autonomous navigation of an unmanned aerial vehicle (UAV). A rule-based system is used for the recognition of visual landmarks such as bridges in aerial views. In principle, rule-based systems can be designed and coded with no training data at hand, but a sound interpretation and utilization of the achieved results needs statistical inference and representative data sets of sufficient coverages. Flying a UAV with an experimental system is expensive, risky, and legally questionable. Therefore, we chose a virtual globe as a camera simulator providing arbitrary amounts of training and test data. The expected positions of landmarks in the aerial views are modeled by mixture models representing inliers, outliers, and intermediate forms which stem from similar structures appearing frequently in the vicinity of landmarks. The parameters of the corresponding likelihood functions are estimated by the Expectation–Maximization method. Using these estimates, we carry out tests and compare the results for heuristic, pessimistic, optimistic, and Bayesian decision rationales. This performance evaluation reveals the superiority of the Bayesian approach.     

Off-line handwritten Chinese character recognition as a compoundBayes decision problem

A handwritten Chinese character off-line recognizer based on contextual vector quantization (CVQ) of every pixel of an unknown character image has been constructed. Each template character is represented by a codebook. When an unknown image is matched against a template character, each pixel of the image is quantized according to the associated codebook by considering not just the feature vector observed at each pixel, but those observed at its neighbors and their quantization as well. Structural information such as stroke counts observed at each pixel are captured to form a cellular feature vector. Supporting a vocabulary of 4616 simplified Chinese characters and alphanumeric and punctuation symbols, the writer-independent recognizer has an average recognition rate of 77.2 percent. Three statistical language models for postprocessing have been studied for their effectiveness in upgrading the recognition rate of the system. Among them, the CVQ-based language model is the most effective one upgrading the recognition rate by 10.4 percent on the average     

The scaling problem in the pattern recognition approach to machine translation

Statistical machine translation (SMT) has proven to be an interesting pattern recognition framework for automatically building machine translations systems from available parallel corpora. In the last few years, research in SMT has been characterized by two significant advances. First, the popularization of the so called phrase-based statistical translation models, which allows to incorporate local contextual information to the translation models. Second, the availability of larger and larger parallel corpora, which are composed of millions of sentence pairs, and tens of millions of running words. Since phrase-based models basically consists in statistical dictionaries of phrase pairs, their estimation from very large corpora is a very costly task that yields a huge number of parameters which are to be stored in memory. The handling of millions of model parameters and a similar number of training samples have become a bottleneck in the field of SMT, as well as in other well-known pattern recognition tasks such as speech recognition or handwritten recognition, just to name a few. In this paper, we propose a general framework that deals with the scaling problem in SMT without introducing significant time overhead by means of the combination of different scaling techniques. This new framework is based on the use of counts instead of probabilities, and on the concept of cache memory.     

A recursive patterns matching model for the dynamic pattern recognition problem

This paper defines a new recursive pattern matching model based on the theory of the systemic functioning of the human brain, called pattern recognition theory of mind, in the context of the dynamic pattern recognition problem. Dynamic patterns are characterized by having properties that change in intervals of time, such as a pedestrian walking or a car running (the negation of a dynamic pattern is a static pattern). Novel contributions of this paper include: (1) Formally develop the concepts of dynamic and static pattern, (2) design a recursive pattern matching model, which exploits the idea of recursivity and time series in the recognition process, and the unbundling/integration of pattern to recognize, and (3) develop strategies of pattern matching from two major orientations: recognition of dynamic patterns oriented by characteristic, or oriented by perception. The model is instantiated in several cases, to analyze its performance.     

Statistical pattern recognition: a review

The primary goal of pattern recognition is supervised or unsupervised classification. Among the various frameworks in which pattern recognition has been traditionally formulated, the statistical approach has been most intensively studied and used in practice. More recently, neural network techniques and methods imported from statistical learning theory have been receiving increasing attention. The design of a recognition system requires careful attention to the following issues: definition of pattern classes, sensing environment, pattern representation, feature extraction and selection, cluster analysis, classifier design and learning, selection of training and test samples, and performance evaluation. In spite of almost 50 years of research and development in this field, the general problem of recognizing complex patterns with arbitrary orientation, location, and scale remains unsolved. New and emerging applications, such as data mining, web searching, retrieval of multimedia data, face recognition, and cursive handwriting recognition, require robust and efficient pattern recognition techniques. The objective of this review paper is to summarize and compare some of the well-known methods used in various stages of a pattern recognition system and identify research topics and applications which are at the forefront of this exciting and challenging field     

Fast pattern recognition using a neurocomputer

This paper presents a new pattern recognition system based on moment invariants using a neurocomputer. The new pattern recognition system consists of a CCD video camera, an image processing system named FDM, a monitor, two stand lights, an NEC PC-9801 microcomputer and a RICOH RN-2000 neurocomputer; these two different types of computers can be considered to constitute an artificial brain. Experimental studies to recognize five dynamic patterns of Japanese chestnuts were performed. From the studies, a high speed of both learning and recognition has been achieved compared with the former pattern recognition system based on the software of artificial neural networks developed by us. This work was presented, in part, at the International Symposium on Artificial Life and Robotics, Oita, Japan, February 18–20, 1996     

Passive stereo range imaging for semi-autonomous land navigation

Many navigation tasks, such as the use of unmanned vehicles for planetary exploration or defense reconnaissance, require onboard range sensors to automatically detect obstacles in the path of a vehicle. Passive ranging via stereo triangulation, or stereo vision, is a very attractive approach to obstacle detection because it is nonemissive, nonmechanical, nonscanning, and compatible with stereographic viewing by human operators. However, several problems have restricted the practicality of stereo ranging in the past, including limitations on the speed, reliability, and generality of existing stereo matching algorithms. This situated has changed, because the Jet Propulsion Laboratory (JPL) recently demonstrated the first semi-autonomous, robotic traverses of natural terrain to use stereo vision for obstacle detection, with all computing onboard the vehicle. This article reviews the main algorithmic paradigms for stereo vision, describes a near realtime stereo vision system developed at JPL, and presents experimental results that demonstrate the emerging practically of stereo vision for obstacle detection in semi-autonomous land navigation. © 1992 John Wiley & Sons, Inc.     

Evolving pattern recognition systems

A hybrid evolutionary learning algorithm is presented that synthesizes a complete multiclass pattern recognition system. The approach uses a multifaceted representation that evolves layers of processing to perform feature extraction from raw input data, select cooperative sets of feature detectors, and assemble a linear classifier that uses the detectors' responses to label targets. The hybrid algorithm, called hybrid evolutionary learning for pattern recognition (HELPR), blends elements of evolutionary programming, genetic programming, and genetic algorithms to perform a search for an effective set of feature detectors. Individual detectors are represented as expressions composed of morphological and arithmetic operations. Starting with a few small random expressions, HELPR expands the number and complexity of the features to produce a recognition system that achieves high accuracy. Results are presented that demonstrate the performance of HELPR-generated recognition systems applied to the task of classification of high-range resolution radar signals.     

Flexible linguistic pattern recognition

“Linguistic” techniques have been developed for well-formed linear strings. This paper describes several extensions of standard linguistic recognizers, to give flexibility in handling loosely connected, noisy, messy patterns.     

On fingerprint pattern recognition

Stochastic languages have been used for fingerprint classification⁽²⁾. In this note we observe that it is not possible to classify the fingerprint patterns in the absence of accurate registration of the fingerprint.Use of features like downward fork, upward fork, end, etc. have been suggested⁽⁴⁾ for fingerprint classification. It is noted that some of the results are either incorrect or incomplete. Limitations of the various algorithms are presented.     

On a recognition problem on cut polytope relaxations

We study the integer recognition problem on cut polytope relaxations. We establish characteristic properties of relaxation points that preclude efficient solution of the problem. We give examples of such points