Showing papers on "Intelligent word recognition published in 1988"

Chinese character recognition: a twenty-five-year retrospective

Abstract: The author assesses the current status of the field and places the problem of Chinese recognition into perspective with other areas of optical character recognition. Early experiments are briefly reviewed, and sources of more up-to-date information, including review articles, are indicated, advances in computer technology are discussed that have had a significant impact on the problem, and a sampling of relatively recent research on the classification of both printed and handprinted ideographs is presented. Included in the discussion are techniques of preprocessing (character location and segmentation) and hierarchical classification. >

A blackboard-based approach to handwritten ZIP code recognition

Abstract: A methodology for recognizing ZIP codes (US postal codes) in handwritten addresses is presented that uses many diverse pattern recognition and image processing algorithms. Given a high-resolution image of a handwritten address block, the solution invokes routines capable of hypothesizing the location of the ZIP code, segmenting and recognizing ZIP code digits, locating and recognizing city and state names, and looking up the results in a dictionary. The control structure is not strictly sequential, but rather in the form of a blackboard architecture that opportunistically invokes routines as needed. An implementation of the methodology is described as well as results with a database of grey-level images of handwritten addresses (taken from live mail in a US Postal Service mail processing facility). Future extensions of the approach are discussed. >

Processes In Human Kanji-word Recognition

Abstract: This paper proposes a human Kanji-word recognition model consisting of two separate processing stages: an early parallel stage and the later serial stage. The model is influenced by word shape knowledge only at the early parallel stage. The model interprets several experimental results- forced-choice identification is more accurate for Kanji characters in Kanji-words than for single Kanji characters in dissimilar alternative sessions, but less accurate in similar alternative sessions. An application of this model to the development of a Kanji-word recognition machine is discussed.

Syntactic and contextual post-processing of handwritten addresses for optical character recognition

Abstract: An optical character recognition (OCR) system is developed for recognizing handwritten and handprinted addresses which include a British postcode written within character boxes. The system makes use of syntactic information concerning postcodes and a postcode database which interacts with the character recognition process to ensure that only valid postcodes are recognized. Postulated valid postcodes are then verified using semantic features of the remainder of the address, to produce a final postcode which both matches the input characters and is compatible with the remainder of the address. Preliminary results are presented to show how the initial stages of syntactic verification can improve character recognition performance. >

Handwritten Character Recognition Adaptable to the Writer.

Abstract: I n this paper, we describe the handwritten character recognition adaptable to the writer. I t is efficient when the specific writer uses the same OCR for many characters. At the early stage, input characters are recognized using genera 1 dictionary, and then the correctly recognized character modify the dictionary to be adaptable to the variation of the characters of the specific writer. Using the adaptable dictionary modif ied by 10 characters/category, the classification rate is improved from 96.8% ( general dictionary ) to 99.5%.

Oblique word forms in visual word recognition

Abstract: It is a common finding in word-recognition experiments that base forms of words are better recognized than oblique forms. It is shown that rather than being due to morphological complexity, this effect is caused by the special status of base forms. However, this status becomes irrelevant when words are presented in syntactic context. A sketch of a model accounting for the data of some recent experiments concludes the paper.

A structural approach to online Chinese character recognition

Abstract: A fuzzy attributed finite-state grammar and corresponding automation are introduced for recognizing line segments and an ordering arrangement is provided according to intrinsic structural properties of Chinese characters. Arrangement is provided. As a result, an online Chinese character recognition system has been implemented. The system consists of a small input tablet interfaced with a IBM-PC/XT/AT which can recognize about seven thousand online handwritten Chinese characters, with some constraints. >

Text recognition system for Japanese documents

Abstract: A text recognition system for Japanese documents is described. The system is composed of a personal computer, which is used as a controller; an image scanner; and a recognition unit. There are four processing stages: text-line segmentation, character segmentation, character recognition, and postprocessing using the Japanese dictionary. Experimental results of the tests for Japanese handwritten technical reports are presented. >

Character recognition of cursive scripts

Abstract: A method for generating the cursive scripts of Arabic-Farsi -Urdu fami ly of languages was developed by Hyder (1). This principle, commonly known as contextual analysis, has now become the universal standard for all input-output devices: e.g. computer terminals, teleprinters, typewri ters, etc. In the present paper the inverse problem is studied and the design of a PROLOG based system is described which recognizes the printed cursive script to determin the constituent characters from which the script had been generated. The procedure described in ( l ) may be considered as a set of inference (production) rules of the type P -* Q, where P is a string of discrete and invariant characters of the alphabet and Q is the string of the corresponding graphemes or character shapes that are context dependent. The printed (cursive) character shape, an element in Q, may have up to six d i f ferent values corresponding to a character, member of the string P. The recognit ion problem may be described as the inverse of the above, that is, for a given Q we determine P, by a set of inference rules of the type Q -~ P, that use pattern matching and unif icat ion in PROLOG. This approach tends to reduce the search space on the average by f ive orders of magnitude. The results, reported in the paper, can be used for the development of an Optical Character Recognition System for these scripts.

Handwritten character recognition dictionary creating system

Abstract: PURPOSE: To efficiently register a necessary writing order in a recognition dictionary by setting the writing order to dictionary strokes to compose a standard character pattern read out from the recognition dictionary under a state in which the strokes of a handwritten character inputted from a tablet correspond to the dictionary strokes according to the writing order of the strokes of the handwritten character. CONSTITUTION: The handwritten strokes are successively outputted one by one from a handwritten pattern input part 16 according to the writing order of the handwritten character pattern inputted from a tablet 14, each dictionary stroke to give the minimum distance is detected based on inter-stroke distance calculation, and an operator is requested to confirm the detected result. When the determination of the writing order for all the dictionary strokes completes, a standard pattern with writing order synthesizing part 22 registers the standard character pattern composed of the set of dictionary strokes with the writing order to compose one character obtained up to that time in a recognition dictionary 10. Thus, the necessary writing order can be efficiently registered in the dictionary. COPYRIGHT: (C)1990,JPO&Japio

Experiments with an n-tuple recogniser for fast 'first try' recognition of unconstrained handwritten symbols

Abstract: The recognition of relatively unconstrained handwritten characters using the n-tuple recognition technique in two application areas is reported. In the case of Electronic Paper, a high resolution flat panel display with a transparent digitizer on its surface, the character set consists of 73 distinct characters and in the case of British postcodes the character set consists of 36 distinct characters. The technique is investigated as a fast method of producing a ranked list of likely classes for each character. The results presented conclude that the n-tuple recognizer is capable of achieving approximately 60% correct first-choice classification on totally unconstrained handwritten characters from any writer and over 90% correct first-choice classification on unconstrained handwritten characters from one writer if the class size is between 15 and 20. In cases where the actual character is not at the top of the ranked list it is usually in a high-ranked position. Recognition performance deteriorates rapidly as the class size exceeds 20. >

A Structural Recognition Algorithm For Handwritten Numerals

Abstract: A new structural recognition method for handwritten numerals is proposed that uses (i) a set of topological pattern primitives, (ii) an adaptive hierachical image segmentation technique and (iii) a tree automaton. Experimental results are also presented.

Character recognizing method

Abstract: PURPOSE: To correct a character, for which a recognition error is generated, by simple processing by executing the communication of a processing progress condition and a recognition candidate character between a recognizing task and a correcting task by using a communicating means. CONSTITUTION: The recognizing task is provided to extract the recognition candidate character by segmenting a recognition object character from a picture, which includes the recognition object character, and collating the character with a dictionary which stores character information. Then, the correcting task is provided to execute the correction and edition of the recognition candidate character. The communication of the processing progress condition and recognition candidate character is executed between the recognizing task and correcting task by using the communicating means. Thus, by using the communicating means with executing recognition processing in the recognizing task or by using a shared memory and the communicating means, the communication of the processing progress condition and recognition candidate character is executed. Then, the correction of the recognition candidate character can be simultaneously executed by the correcting task. COPYRIGHT: (C)1990,JPO&Japio

A decision making method and its application in unconstrained handwritten character recognition

Abstract: A decision making method capable of dealing with the problems faced in real-life applications is developed. Its performance on 2759 totally unconstrained handwritten numerals is measured. The estimated recognition reliability of this method for the training set samples is 99.8% and for the testing set is 99.06%.

Flexible recognition algorithms and flexible OCR system

Abstract: An optical handwritten numeral recognition system is presented which uses a G3 facsimile transceiver (fax) as the input device. Flexible recognition algorithms perform raster scanning of the Huffman code produced by the fax only once. The hanger-chain algorithms separate handwritten numerals and obtain the first features at the same time. All the recognition software is coded in Turbo C language. The system recognition speed is from 10 to 100 characters per second, and the recognition rate is from 95% to 99.5% depending on the sample. >

Omnifont character recognition based on fast feature vectorization

Abstract: This paper presents major achievements made towards the development of a high-speed optical character recognition (OCR) workstation for characters of various fonts and sizes. The system is based upon an efficient feature extraction concept centred around an edge-vectorization technique. The resulting edges are mapped into a feature space from where a binary feature vector is built and subsequently fed to a standard statistical Bayesian classifier. The technique has been demonstrated on an IBM-PC/XT (without coprocessor) to operate at least 25 times the speed of conventional OCR techniques, achieving a 100% recognition rate with learned characters and 87% with unlearned.