Context in word recognition

Relatively low character error rates can often lead to prohibitive levels of word error rates. This paper examines several techniques for integrating an independent contextual postprocessor (CPP) into a full classification system. Using positional binary n-grams the CPP can correct many errors directly. In those cases where the correction process leads to ambiguity, the CPP can direct additional processing. Experimental results demonstrate that almost all of the derived improvement results from CPP-directed reclassification. This only requires that the CPP have the classifier likelihood fed forward to it. Therefore, a standardized CPP can be built independently of the rest of the classification system. An initial 45% word error rate is reduced to about a 2% word error rate and a 1% reject rate. Presence of a dictionary allows these figures to be reduced even further.     

Testing the performance of spoken dialogue systems by means of an artificially simulated user

This paper proposes a new technique to test the performance of spoken dialogue systems by artificially simulating the behaviour of three types of user (very cooperative, cooperative and not very cooperative) interacting with a system by means of spoken dialogues. Experiments using the technique were carried out to test the performance of a previously developed dialogue system designed for the fast-food domain and working with two kinds of language model for automatic speech recognition: one based on 17 prompt-dependent language models, and the other based on one prompt-independent language model. The use of the simulated user enables the identification of problems relating to the speech recognition, spoken language understanding, and dialogue management components of the system. In particular, in these experiments problems were encountered with the recognition and understanding of postal codes and addresses and with the lengthy sequences of repetitive confirmation turns required to correct these errors. By employing a simulated user in a range of different experimental conditions sufficient data can be generated to support a systematic analysis of potential problems and to enable fine-grained tuning of the system.     

The artificial intelligence approach to pattern recognition—a perspective and an overview

The present paper reviews the techniques for automated extraction of information from signals. The techniques may be classified broadly into two categories—the conventional pattern recognition approach and the artificial intelligence (AI) based approach. The conventional approach comprises two methodologies—statistical and structural. The paper reviews salient issues in the application of conventional techniques for extraction of information. The systems that use the artificial intelligence approach are characterized with respect to three key properties. The basic differences between the approaches and the computational aspects are reviewed. Current trends in the use of the AI approach are indicated. Some key ideas in current literature are reviewed.     

隐喻逻辑的研究进展

讨论了当今隐喻逻辑这一研究领域中存在的一些主要问题.对这一领域的一系列思想进行评述.得出的结论是,隐喻逻辑的建立离不开对一些隐喻赖于存在的根本逻辑的建立,如相似和理解的逻辑描述,以便使隐喻逻辑更加严密和完整.     

Texture analysis of aerial photographs

Different image textures manifest themselves by dissimilarity in both the property values and the spatial interrelationships of their component texture primitives. We use this fact in a texture discrimination system.An image is first segmented into closed regions called units. Then, a set of properties is calculated for each of the units. The units along with their respective properties constitute the primitives.The discrimination between texture categories has two parts: the training phase and the classification phase. The primitives and the relationships which are obtained from representative training images are used to develop criteria for the classification phase. During classification, the primitives of the image under test are first used to assign a unit to one of several cluster types. Then, each primitive is assigned to the most likely texture class given its cluster type and the cluster types of its spatially adjacent neighbors.The method is used on three images: a noisy checkerboard, a simulated texture and an aerial photograph.     

A survey on the use of pattern recognition methods for abstraction, indexing and retrieval of images and video

The need for content-based access to image and video information from media archives has captured the attention of researchers in recent years. Research efforts have led to the development of methods that provide access to image and video data. These methods have their roots in pattern recognition. The methods are used to determine the similarity in the visual information content extracted from low level features. These features are then clustered for generation of database indices. This paper presents a comprehensive survey on the use of these pattern recognition methods which enable image and video retrieval by content.     

Correspondence analysis in the context of pattern recognition

Correspondence analysis is a multivariate technique that is used for the analysis of discrete data. Its formalism is presented, together with the application to mixed data. Some properties are enumerated. Its use in pattern recognition is discussed both in the context of classification of single objects and samples (collection of objects).     

Advances in software engineering and their relations to pattern recognition and image processing

In recent years ‘software engineering’ has emerged as a discipline of programming. It includes the conceptualization, design, implementation, testing and modification of software systems. Related issues are languages, standards, distributed and parallel processing, and total programming environments. The fact that more than 80 percent of system development costs were in software rather than hardware helps one appreciate the importance of any effort to understand and enhance the software production process. Many pattern recognition projects involve fairly large software efforts. It makes sense not only for researchers to make use of the latest software tools and methologies but also to anticipate future changes.     

A scene description method using three-dimensional information

This paper presents a method for describing scenes with polyhedra and curved objects from three-dimensional data obtained by a range finder. A scene is divided into many surface elements consisting of several data points. The surface elements are merged together into regions. The regions are classified into three classes: plane, curved and undefined. The program extends the curved regions by merging adjacent curved and undefined regions. Thus the scene is described by plane regions and smoothly curved regions, which might be useful for the recognition of the objects. From the results obtained so far the program seems to achieve the desired goals.     

A parallel processing model for the study of possible pattern recognition mechanisms in the brain

An electronic model, based on parallel processing concept and able to identify a series of patterns without requiring sophisticated technology, is presented. A receptive part transduces the pattern images into electrical signals and a processing part identifies them. The basic element of the processing part is a “pattern detecting” unit which produces a maximal response for one specific pattern presented on the receptive matrix. A set of 11 “pattern detecting” units allows the model to recognize simple patterns under various conditions and through a neural-like organization, to identify them regardless of their position or size.     

Intelligent compilation of patent summaries using machine learning and natural language processing techniques

Patents are a type of intellectual property with ownership and monopolistic rights that are publicly accessible published documents, often with illustrations, registered by governments and international organizations. The registration allows people familiar with the domain to understand how to re-create the new and useful invention but restricts the manufacturing unless the owner licenses or enters into a legal agreement to sell ownership of the patent. Patents reward the costly research and development efforts of inventors while spreading new knowledge and accelerating innovation. This research uses artificial intelligence natural language processing, deep learning techniques and machine learning algorithms to extract the essential knowledge of patent documents within a given domain as a means to evaluate their worth and technical advantage. Manual patent abstraction is a time consuming, labor intensive, and subjective process which becomes cost and outcome ineffective as the size of the patent knowledge domain increases. This research develops an intelligent patent summarization methodology using artificial intelligence machine learning approaches to allow patent domains of extremely large sizes to be effectively and objectively summarized, especially for cases where the cost and time requirements of manual summarization is infeasible. The system learns to automatically summarize patent documents with natural language texts for any given technical domain. The machine learning solution identifies technical key terminologies (words, phrases, and sentences) in the context of the semantic relationships among training patents and corresponding summaries as the core of the summarization system. To ensure the high performance of the proposed methodology, ROUGE metrics are used to evaluate precision, recall, accuracy, and consistency of knowledge generated by the summarization system. The Smart machinery technologies domain, under the sub-domains of control intelligence, sensor intelligence and intelligent decision-making provide the case studies for the patent summarization system training. The cases use 1708 training pairs of patents and summaries while testing uses 30 randomly selected patents. The case implementation and verification have shown the summary reports achieve 90% and 84% average precision and recall ratios respectively.     

A syntactic procedure for the recognition of glottal pulses in continuous speech

Speech prosody contains an important structural information for performing speech analysis and for extracting syntactic nuclei from spoken sentences. This paper describes a procedure based on a multichannel system of epoch filters for recognizing the pulses of glottal chord vibrations by an analysis of the speech waveform. Recognition is performed by a stochastic finite state automaton automatically inferred after experiments.     

Combined pattern search optimization of feature extraction and classification parameters in facial recognition

Constantly, the assumption is made that there is an independent contribution of the individual feature extraction and classifier parameters to the recognition performance. In our approach, the problems of feature extraction and classifier design are viewed together as a single matter of estimating the optimal parameters from limited data. We propose, for the problem of facial recognition, a combination between an Interest Operator based feature extraction technique and a k-NN statistical classifier having the parameters determined using a pattern search based optimization technique. This approach enables us to achieve both higher classification accuracy and faster processing time.     

C-calculus: An elementary approach to some problems in pattern recognition

This work utilizes the concept of “Composite set” (C-set) and of the related C-calculus to study some standard problems of pattern analysis and general processing of signals. After some basic definitions and notations about composite sets are briefly stipulated, it is shown how a family of C-sets can be associated with a digitized picture. Each element in the family conveys partial information about the picture itself, yet it is possible to combine the various contributions from each C-set in such a way as to completely retrieve the image. Conditions that guarantee such “convergence” are theoretically investigated: the cases of nonconvergence are also proved to be of some interest.

C-calculus is concretely applied to the extraction of significant regions in a digitized picture, of contours, etc. An application to texture discrimination and analysis is also outlined.     

Pattern recognition processing in underwater acoustics

Pattern recognition application to underwater acoustics is a relatively less explored area, even considering some reports which may not be available in the public domain. In this paper three problem areas of pattern recognition processing in underwater acoustics are reported. The first problem is three-dimensional target motion analysis. A reference map is generated and partitioned into a number of cells corresponding to possible target locations. The pattern matching idea is then used to estimate the target range, velocity and bearing. The second problem is concerned with multipath ranging. An image processing technique is used to extract the significant tracks from the correlograms to provide a continuous estimate of time delay or range under a multipath environment. The third problem deals with submarine transient signals. A spectral analysis is made to determine spectral features for detection and event classification. The use of entropy distance measure for waveform segmentation is then examined. Major computer results are presented along with a discussion of unresolved issues for each problem area.     

Recent progress to formal approach of pattern recognition and scene analysis

The P.R. techniques and programs may currently be considered as a “bag of tricks”, justified by their experimental results. Formalization and theory have been lagging. Propositions are made for a formal P.R. language, similar to the language of logic, which could be used to describe precisely the P.R. algorithms. Interpretations of the terms of this language have to be made in other languages: natural, machine or programming languages. A P.R. function should be constructive, i.e. find its interpretation in an algorithm or program. A training set is far from enough to define such a function. Properties of the experimental domain should guide the P.R. specialist. They are examined and discussed, in an effort to find some unity in the P.R. techniques.     

The application of advanced pattern recognition techniques for the discrimination between earthquakes and nuclear detonations

This paper presents a study of several features thought to be useful in the discrimination between the seismic signatures generated by nuclear detonations and those generated by earthquakes. All signatures were recorded at teleseismic distances. Two new features are introduced; the adaptive-static and the adaptive-dynamic spectral ratios. The data base used in this study consisted of approximately 250 earthquakes and 150 nuclear detonations. The signatures were transformed into feature sets by means of third moments of frequency as well as complexity, normal-static, adaptive-static, normal-dynamic, and adaptive-dynamic spectral ratios. Exponential potential functions were used to generate the classification hyperplanes. The data base was randomly divided into learning and evaluation subsets. A 99% classification was achieved for the learning subset and a 97% classification was obtained for the evaluation subset.     

A comparison study of pattern recognition algorithms implemented on a microcontroller for use in an electronic tongue for monitoring drinking waters

A portable electronic tongue has been developed using an array of eighteen thick-film electrodes of different materials forming a multi-electrode array. A microcontroller is used to implement the pattern recognition. The classification of drinking waters is carried out by a Microchip PIC18F4550 micro-controller and is based on neural networks algorithms. These algorithm are initially trained with the multi-electrode array on a Personal Computer (PC) using several samples of waters (still, sparkling and tap) to obtain the optimum architecture of the networks. Once it is trained, the computed data are programmed into the microcontroller, which then gives the water classification directly for new unknown water samples. A comparative study between a Fuzzy ARTMAP, a Multi-Layer Feed-Forward network (MLFF) and a Linear Discriminant Analysis (LDA) has been done in order to obtain the best implementation on a microcontroller.