String Recognition Research Articles

Development of an expert system for recognizing handwritten dollar amounts

AbstractThis paper presents an expert system for automatically recognizing handwritten characters of dollar amounts on personal checks. The dollar amounts composed of numerals and marks—, “X X,” etc., are freely handwritten in various formats that individuals favor, inside the amount field. Then frequently there is contact between not only horizontally adjacent characters but also vertically adjacent ones. Recognition of such unconstrained character strings handwritten by the public whose formats are undecided beforehand is a problem in optical character recognition.The system developed in this paper has the ability to segment the dollar amount image into individual characters automatically and to recognize them accurately. The recognition process of the dollar amount consists of the following steps: divide a digitized image into groups, called a “block,” of line elements connected with one another; give attributes to each block by using the knowledge on arrangement patterns of characters; and segment the block into individual characters according to the attributes and recognize them. It is suggested that this system will enable the saving of considerable operator labor in check processing.

On the applicability of VQ codebooks to connected digit recognition

Connected digit recognition is a problem that has received a lot of attention over the past several years because of its importance in providing speech recognition services (e.g., catalog ordering, credit card entry, all digit dialing of telephone numbers, etc.). Although a number of systems have been described that provide very high string accuracy on a standard database of connected digits (i.e., the TI database), most of these systems require a great deal of computation to provide high performance. Most recently, there has been a renewed interest in connected digit recognition systems based on discrete density models using VQ codebooks (e.g., the work of Normandin and colleagues at CRIM in Montreal) where the computation is significantly lower than that required for continuous density models, and the robustness to variations in talkers, background, microphones, etc., has the potential to be high. In this study, the effect of multiple codebooks, multiple models, and codebook weighting on the performance of a standard hidden Markov model recognizer using the TI connected digits database is examined. It is shown that a single codebook can provide high string recognition accuracy, and that multiple codebooks provide accuracy comparable to that of continuous density models.

The utilization of fuzzy sets in the recognition of imperfect strings

Abstract The intent of this paper is to present a new method in imperfect string recognition, when classical error detection algorithms are employed on a given grammar. First we will describe the classic approach and then we will introduce the new approach. We present examples that describe the use of fuzzy sets in the recognition of imperfect strings.

A speaker-independent connected digit recognition system concatenating statistically discriminated words

A recognition system for connected digits, which uses a statistical classifier to identify words in speaker-independent continuous speech, is described. The system uses the multiple similarity method, a statistical pattern recognition technique. For evaluating word strings, the system uses a scoring method that is independent of the number of words in the strings. It is derived from the a posteriori probability that a subinterval corresponds to a correct word position, giving a word similarity value. The system evaluates a word string using dynamic programming and a parallel search procedure. Experiments for the contextual effect of the training data set, for validation of the search algorithm, and for a large quantity of unspecified speakers including 40 males and 40 females were performed. For connected digits (unknown word lengths test), the string recognition rates were 90.1%-95.1% for two, three, or four connected digits, where the equivalent word (digit) rates were 97.4%-98.4%.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Recognition of letters in lateral printed strings using a three-layered BP model with feedback connections

Although there is a simple and fast method to separate and classify letters in lateral printed strings by regarding a valley of the projection of the original image as a gap of letters, this method cannot be applied to cases where adjacent letters are connected or large clutter is involved [1]. This paper proposes a method which separates and recognizes each letter even in such cases as described in the foregoing, using a three-layered BP model with feedback connections. The recognition experiments using the proposed method were carried out with lateral strings of five printed letters (capital A to E) binarized through a CCD camera. The results show that the method is superior to conventional methods when applied to a letter string recognition system used in the process of factory automation.

Prosodic information in telephone digit strings

Intonation of naturally produced telephone digit strings appears to conform to particular patterns. Concatenating strings of digits produced with “default” intonation highlights the importance of maintaining appropriate intonation patterns for intelligible, natural‐sounding speech. The purpose of this experiment was to quantify the acoustic events resulting in natural‐sounding telephone number digit strings. Acoustic features of approximately 1000 digits produced in the context of telephone digit strings were measured. The data were gathered at MIT by presenting volunteer speakers with lists of seven‐digit numbers to read. Shifts in fundamental frequency of the vowel, changes in overall energy, and digit duration as a function of position in the digit string were calculated. These results can prove useful for speech recognition and generation. For automatic speech recognition of digit strings, prosodic contours provide cues about word boundary locations and serve as a source of information for error detection and correction. When concatenating stored speech to produce natural sounding digit strings, these findings provide guidance as to how many templates need to be created. Finally, for text‐to‐speech synthesis, these results suggest the number of rules necessary to model natural‐sounding digit strings.

Method and apparatus for continuous word string recognition

A speech recognition method and apparatus for recognizing word strings in a continuous audio signal are disclosed. The word strings are made up of a plurality of elements, and each element, generally a word, is represented by an element template defined by a plurality of target patterns. Each target pattern is represented by a plurality of statistics describing the expected behavior of a group of spectra selected from plural short-term spectra generated by processing of the incoming audio. Each target pattern has associated therewith at least one required dwell time position and at least one optional dwell time position. The number of required dwell time positions and the sum of the required and optional dwell time positions define, in effect, the limits of a time interval during which a given target pattern can be said to match an incoming sequence of frame patterns. The incoming audio spectra are processed to enhance the separation between the spectral pattern classes during later analysis. The processed audio spectra are grouped into multi-frame spectral patterns and are compared, using likelihood statistics, with the target patterns of the element templates. Each multi-frame pattern input, which inputs occur at a frame rate which requires each keyword target pattern to correspond to at least two of the multi-frame patterns, is forced to contribute to each of a plurality of pattern scores as represented by the element templates. The contributions of said multi-frame pattern inputs to said pattern scores is controlled, in part, by said required and optional dwell time constraints. A concatenation technique is employed, using dynamic programming techniques, to determine the correct identity of the word string.

An automatic circuit diagram reader with loop-structure-based symbol recognition

A high-performance logic circuit diagram reader was developed for VLSI-CAD data input. Almost all logic circuit symbols include one or more loop structures. A description is given of an efficient method for recognition of these loop-structured symbols. The proposed method consists of two processes: symbol segmentation and symbol identification. Symbol identification is achieved by a powerful hybrid method which uses heuristics to mediate between template matching and feature extraction. The entire symbol recognition process is carried out under a decision-tree control strategy. The entire recognition system for circuit diagrams is briefly explained, including character string recognition and connecting line analysis. >

Network-based connected digit recognition

A system for speaker-independent connected digit recognition is described in which explicit acoustic-phonetic features and constraints play a significant role. The digit vocabulary is modeled using a finite-state pronunciation network whose branches correspond to meaningful acoustic-phonetic units. Each branch is associated with an acoustic pattern matcher which employs a combination of whole-spectrum and feature-based metrics. The system has been evaluated using 17 000 utterances from the Texas Instruments (TI) multidialect, connected digits database. The best configurations of the recognizer achieve string recognition accuracies of approximately 96 and 97 percent when the length of the input string is unknown and known, respectively, and when different talkers are used for training and testing.

Recognizing Melodies: A Dynamic Interpretation

Two experiments explore hypotheses about rhythm and contour in recognition of simple pitch strings (melodies). Target melodies that differed with respect to pitch relationships (interval and contour pitch differences) and rhythm, were presented to ordinary listeners who were told to learn the melodies (Phase I). In a subsequent recognition test (Phase II), listeners had to recognize these same target melodies although they were transposed to a different musical key. In recognition, target melodies appeared in the original rhythm or in new rhythms that simulated some pause properties of the original rhythm. Target melodies were interspersed with decoy melodies that either preserved the pitch contour of targets or did not; all appeared in the original rhythm and in new rhythms. Results indicated that a new rhythmic context lowered recognizability of target melodies, and that decoys were most confusing when they possessed the same “dynamic shape” (contour-plus-rhythm) as targets (Experiment 1). Also, target recognition improved with Phase I familiarity (Experiment 2), although rhythmic shifts remained detrimental across levels of target familiarity. Confusions based on “dynamic shape” accounted for a relatively high proportion of errors where familiarity with targets is low. Findings were interpreted in terms of a theory of context-sensitive dynamic attending in which remembering is assumed to involve recapitulation of the original rhythmical activities involved in attending to melodies.

Visual knowledge underlying letter perception: Font-specific, schematic tuning.

The representation of visual information about letters is proposed to be highly systematic, involving not only abstract information that is invariant across type faces (or fonts), but also a number of parameters whose values are determined by the current font. The system exploits regularities that are characteristic of letters and fonts by becoming tuned to the details of the font. This should result in efficient letter perception when the stimuli are regular (when all of the letters are of a consistent font), but not when the stimuli are irregular (when the letters are from a variety of fonts). The prediction of faster processing with a regular font, as compared with a mixed font, was examined in three experiments requiring the recognition of four-letter strings. Experiment 1 confirmed the prediction, and Experiment 2 replicated the effect with the number of "features" equated across conditions. Experiment 3 showed that the disadvantage for a mixture of fonts is related to how much the representational system must be adjusted to handle the different fonts.

Generation of efficient LALR parsers for regular right part grammars

A method for building small fast LALR parsers for regular right part grammars is given. No grammar transformation is required. No extra state of the LALR parser for the recognition of strings generated by a right part is required. At some reduce states the parser may refer to lookback states (states in which the parser may be restarted after the reduction). An optimizing algorithm to reduce these references is also given.

Median strings

The concept of ‘median string’ over a set of strings is introduced in this paper. The set median is defined to be that element in the set of strings which has the smallest sum of distances (with respect to an arbitrary measure) from the other elements, and the (generalized) median is a hypothetical, artificially constructed element which has the smallest sum of distances from all the elements of the given set. The applicability of the median strings in the recognition of garbled strings is demonstrated, where they mostly yield a higher accuracy than the multiple-similarity and the k-nearest-neighbour methods.

Typing letter strings varying in orthographic structure

Subjects typed six-letter strings varying in orthographic structure. Lexical status, word frequency, position-sensitive log bigram frequency, and regularity of letter sequencing were systematically varied. Cumulative reaction times (RTs) of the keystrokes were adequately described by a linear function of letter position in the test string. Overall, words were typed faster than nonwords, and regular strings faster than irregular strings. Although the effect of log bigram frequency was not significant, this variable interacted with regularity and word frequency. Post hoc analyses of performance on each of the 200 letter strings revealed significant effects of the number of irregularities, log bigram frequency, and log word frequency. Transition times between successive keystrokes were significantly longer for illegal than for legal letter transitions. These results are similar to previous findings on the role of orthographic structure in the perceptual recognition of letter strings and provide a more complete analysis of context effects in typing.

Parallel string acceptance using lattice graphs

This paper describes a method of string acceptance with respect to a given grammar G by repeatedly applying the rules of G in parallel. The process differs from previous work in parallel string recognition in that when a rule, say α→ β, is applied, β is not replaced by α; rather, a ‘string bypass’, consisting of α, is created that joins the left context of β to its right context. Thus, the process creates a ‘lattice graph’ in which any directed path from the least element to the greatest element is a string derivable (in reverse) by a sequence of rewritings of σ. For context-free grammars, this process does not lead to a combinatorial explosion, and could thus be used to parse strings very rapidly if suitable parallel hardware were available.

On-line recognition of handprinted Kanji characters

Difficulties in Kanji (Chinese character) recognition stem from its large character set (about 5000 characters) and the large number of strokes (up to about sixty) in each character. The paper describes a preliminary approach to this Kanji recognition problem. In the present method, a handprinted Kanji character is coded into a symbol string using the binary relation between stroke and reference zone. Two symbol string recognition methods are proposed and investigated; the direct matching recognition (DMR) method and the unit structure recognition (USR) method. The DMR method worked efficiently for characters which have up to five strokes. The USR method represents Kanji characters with a structural unit combination. This method worked efficiently for multi-stroke characters and greatly reduced dictionary update labor.

Cellular graph automata. II. graph and subgraph isomorphism, graph structure recognition

This paper deals with cellular automata in which the intercell connections define a graph of bounded degree. It discusses acceptance tasks that involve the detection of graph or subgraph isomorphism in time proportional to the diameter of the given graph. In some of the algorithms, special assumptions are made about the “homogeneity” of the graph; these assumptions hold for many important classes of graphs, including trees and arrays. The paper also examines types of graph structures that can be recognized by these automata. Diameter-time algorithms are presented for the recognition of cycles, strings, trees, cliques, rectangular and square arrays, Eulerian graphs, bipartite and complete bipartite graphs, stars, and wheels. The recognition of planar graphs is also discussed.

The role of orthographic and phonotactic rules in perceiving letter patterns.

Three experiments examined the role of orthographic and phonotactic rules in the tachistoscopic recognition of letter strings. Experiment 1 showed that the presence of a vowel or multiletter spelling patterns facilitates perceptual accuracy. To account for these results a model was proposed in which an input string is first parsed into syllablelike units, which are then recorded into speech. It was demonstrated that the perceptual accuracy for a string is correlated with the number of recoding steps needed to convert that string into speech. Experiment 2 further demonstrated that this recoding process can predict perceptibility differences among strings with varying numbers of phonotactic violations, and Experiment 3 assessed some of the specific assumptions of the recoding process.

The role of orthographic and phonotactic rules in perceiving letter patterns.

Three experiments examined the role of orthographic and phonotactic rules in the tachistoscopic recognition of letter strings. Experiment 1 showed that the presence of a vowel or multiletter spelling patterns facilitates perceptual accuracy. To account for these results a model was proposed in which an input string is first parsed into syllablelike units, which are then recorded into speech. It was demonstrated that the perceptual accuracy for a string is correlated with the number of recoding steps needed to convert that string into speech. Experiment 2 further demonstrated that this recoding process can predict perceptibility differences among strings with varying numbers of phonotactic violations, and Experiment 3 assessed some of the specific assumptions of the recoding process.

Expectancy as a determinant of functional units in perceptual recognition

This experiment provides support for (1) Gibson et al.'s (1962) hypothesis that either an individual letter or a spelling pattern (i.e., a cluster of graphemes in a given linguistic environment which has an invariant pronunciation according to the rules of English) can serve as a functional unit in the tachistoscopic recognition of letter strings, and (2) Neisser's (1967) hypothesis that S's expectancy determines which one of these functional units is used. On each of 16 trials a letter string was presented tachistoscopically followed immediately by a two-choice recognition test of one of the letters. On the first 14 trials half the Ss expected and received letter strings containing spelling patterns (SPs) while the remaining Ss expected and received letter strings containing only unrelated letters. The former group of Ss performed significantly better than the latter group on these 14 trials. On Trial 15 S's expectancy was again confirmed but on Trial 16 S was presented with a letter string from the unexpected class (SPs or unrelated letters). The results from the last two trials showed that letter strings containing SPs were recognized more accurately than unrelated letters when S expected SPs but not when he expected unrelated letters.