By Topic

Frontiers in Handwriting Recognition, 2002. Proceedings. Eighth International Workshop on

Date 6-8 Aug. 2002

Filter Results

Displaying Results 1 - 25 of 88
  • Proceedings Eighth International Workshop on Frontiers in Handwriting Recognition [front matter]

    Save to Project icon | Request Permissions | PDF file iconPDF (295 KB)  
    Freely Available from IEEE
  • Author index

    Page(s): 529 - 530
    Save to Project icon | Request Permissions | PDF file iconPDF (185 KB)  
    Freely Available from IEEE
  • MOrpho-LEXical analysis for correcting OCR-generated Arabic words (MOLEX)

    Page(s): 461 - 466
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (1031 KB)  

    In this paper we present a contextual-based method for correcting Arabic words generated by OCR systems. This technique operates as a post-processor and it wants to be universal. It corrects substitution and rejection errors. The Arabic language properties are very useful in morpho-lexical analysis and therefore they are strongly exploited in the development of the method. The substitution errors, the most frequently committed ones by the OCR systems, are rewritten in production rules to be used by a rule-based system for correcting Arabic words. The first version of the developed method operates only at the morpho-lexical level, the extension to the other levels of language analysis is considered in perspectives. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Vind(x): using the user through cooperative annotation

    Page(s): 221 - 226
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (361 KB) |  | HTML iconHTML  

    In this paper, the image retrieval system Vind(x) is described. The architecture of the system and first user-experiences are reported. Using Vind(x), users on the Internet may cooperatively annotate objects in paintings by use of the pen or mouse. The collected data can be searched through query-by-drawing techniques, but can also serve as an (ever-growing) training and benchmark set for the development of automated image retrieval systems of the future. Several other examples of cooperative annotation are presented in order to underline the importance of this concept for the design of pattern recognition systems and the labeling of large quantities of scanned documents or online data. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A handwritten character recognition method based on unconstrained elastic matching and eigen-deformations

    Page(s): 72 - 77
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (329 KB) |  | HTML iconHTML  

    A fast elastic matching based handwritten character recognition method is investigated. In the method, an unconstrained elastic matching technique, where the matching is optimized locally and individually on each pixel, is utilized together with its a posteriori evaluation based on the eigen-deformations of handwritten characters. Our experimental results show that high recognition rates can be attained by the present method with feasible computations. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Signature verification: benefits of multiple tries

    Page(s): 424 - 427
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (244 KB) |  | HTML iconHTML  

    This method describes the advantages of a signature verification or any other biometric methodology having multiple tries. Traditional verification techniques acquire a test sample and compare with a model and a decision is made as to whether the test sample is genuine or a forgery. While it is clear that allowing a second try will reduce the false rejects, it will also increase the false accepts. However, from our experiments, the increase in false accepts was small compared to the dramatic reduction in false rejects. At the 1% false rejection (FR) rate, the addition of the second try reduced the false acceptance (FA) rate from 4.7% to 1.3%. So, in applications where very low FR is required, allowing the signer a second try appears to be a good option. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Hidden loop recovery for handwriting recognition

    Page(s): 375 - 380
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (439 KB) |  | HTML iconHTML  

    One significant challenge in the recognition of off-line handwriting is in the interpretation of loop structures. Although this information is readily available in online representation, close proximity of strokes often merges their centers making them difficult to identify. In this paper a novel approach to the recovery of hidden loops in off-line scanned document images is presented. The proposed algorithm seeks blobs that resemble truncated ellipses. We use a sophisticated form analysis method based on mutual distance measurements between the two sides of a symmetric shape. The experimental results are compared with the ground truth of the online representations of each off-line word image. More than 86% percent of the meaningful loops are handled correctly. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Automated assessment: how confident are we?

    Page(s): 419 - 423
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (277 KB) |  | HTML iconHTML  

    This paper highlights the research issues associated with the automated assessment of handwritten scripts and introduces the theoretical scoring confidence. Using this concept, in a 3 word response environment, we prove that it is theoretically possible to achieve a scoring confidence greater than 98% using recognition rates as low as 81% to produce actual response yields of 50%. These results are verified by experiment. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Comparing on-line recognition of Japanese and western script in preparation for recognizing multi-language documents

    Page(s): 84 - 89
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (276 KB) |  | HTML iconHTML  

    Recognizers supporting multiple languages and writing systems are becoming increasingly important due to the international success of pen-based systems. The main intention of the paper is to improve our understanding of the differences and similarities between Japanese and western handwriting recognition. Knowing the common techniques is important for developing compact and powerful multi-language recognizers with integrated modules for both writing systems. In particular, the simultaneous recognition of western and Japanese handwriting in multi-language documents requires methods suitable for both writing systems. Though the Japanese and western writing systems are completely different, we present many similar recognition techniques facilitating an integration of processing steps. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Handprinted Hiragana recognition using support vector machines

    Page(s): 55 - 60
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (310 KB) |  | HTML iconHTML  

    Describes a method to improve the cumulative recognition rates of pattern recognition using a decision directed acyclic graph (DDAG) based on support vector machines (SVM). Though the original DDAG has high level of performance and its execution speed is fast, it does not consider the so-called cumulative recognition rate. We construct a DDAG which can incorporate the cumulative recognition rate. As a result of our experiment for handprinted Hiragana characters in JEITA-HP, the cumulative recognition rate is improved and its execution time is almost the same as the original DDAG and 30 times faster than the Max Win Algorithm which is one of the famous recognition methods using support vector machines for a multi-class problem. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A new warping technique for normalizing likelihood of multiple classifiers and its effectiveness in combined on-line/off-line japanese character recognition

    Page(s): 177 - 182
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (523 KB) |  | HTML iconHTML  

    We propose a technique for normalizing likelihood of multiple classifiers prior to their combination. Our technique takes classifier-specific likelihood characteristics into account and maps them to a common, ideal characteristic allowing fair combination under arbitrary combination schemes. For each classifier, a simple warping process aligns the likelihood with the accumulated recognition rate, so that recognition rate becomes a uniformly increasing function of likelihood. For combining normalized likelihood values, we investigate several elementary combination rules, such as sum-rule or max-rule. We achieved a significant performance gain of more than five percent, compared to the best single recognition rate, showing both the effectiveness of our method for classifier combination and the benefit of combining on-line Japanese character recognition with stroke order and stroke number independent off-line recognition. Moreover, our experiments provide additional empirical evidence for the good performance of the sum rule in comparison with other elementary combination rules, as has already been observed by other research groups. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Classifying isogenous fields

    Page(s): 41 - 46
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (303 KB) |  | HTML iconHTML  

    Classifiers that utilize style context in co-occurring patterns increase recognition accuracy. When patterns occur as long isogenous fields, this gain should increase unless negated by parameter estimation errors that increase with field length. We show that our method achieves higher accuracy with longer input fields because it can be trained accurately We also present some ongoing work on simple heuristics to reduce computational complexity of the scheme. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Integrated segmentation and recognition of handwritten numerals: comparison of classification algorithms

    Page(s): 303 - 308
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (380 KB)  

    In integrated segmentation and recognition (ISR) of handwritten character strings, the underlying classifier is desired to be accurate in character classification and resistant to non-character patterns (also called garbage or outliers). This paper compares the performance of a number of statistical and neural classifiers in ISR. Each classifier has some variations depending on learning method: maximum likelihood estimation (MLE), discriminative learning (DL) under the minimum square error (MSE) or minimum classification error (MCE) criterion, or enhanced DL (EDL) with outlier samples. A heuristic pre-segmentation method is proposed to generate candidate cuts and character patterns. The methods were tested on the 5-digit Zip code images in CEDAR CDROM-1. The results show that training with outliers is crucial for neural classifiers in ISR. The best result was given by the learning quadratic discriminant function (LQDF) classifier. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • On line signature verification: Fusion of a Hidden Markov Model and a neural network via a support vector machine

    Page(s): 253 - 258
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (327 KB) |  | HTML iconHTML  

    We propose in this work to perform on-line signature verification by the fusion of two complementary verification modules. The first one considers a signature as a sequence of points and models the genuine signatures of a given signer by a Hidden Markov Model (HMM). Forgeries are used to compute a decision threshold. In the second module, global parameters of a signature are the inputs of a two-classes neural network trained for each signer on both the genuine and "other" signatures (genuine signatures of other signers). Fusion of the scores given by these two experts through a Support Vector Machine (SVM), allows improving the results over those of each module, on Philips' Database. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Affine alignment for stroke classification

    Page(s): 381 - 386
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (346 KB) |  | HTML iconHTML  

    We propose a stroke classification method based on affine alignment, appropriate for online recognition of mathematical handwriting. The method, essentially linear is simple and computationally efficient. The modeling limitations of the affine group are overcome by choosing adequate error functions and by performing alignment with respect to interpolated prototypes. So, moderate nonlinear transformations are tolerated, making the approach invariant to a wide range of handwriting deformations. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • "Schurmann-polynomials - roots and offsprings": Their impact on today's pattern recognition

    Page(s): 3 - 9
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (265 KB) |  | HTML iconHTML  

    Jurgen Schurmann died on January 19, 2001, much too early, at the age of 66 years. He was a grand pioneer in the field of pattern recognition and a tireless source and trigger of sophisticated theoretical ideas and their transformations into high performance recognition products which are spread all over the world. For this reason, the programme committee of the 2002 "International Workshop on Frontiers of Handwriting Recognition" (IWFHR-8) has decided to dedicate this event to this extraordinary senior scientist of pattern recognition. In honour and in memory of this vivid, dynamic and creative man, we want to reflect the essence of his oeuvre; gained during a life-long quest to find a way - as he would put it - "from pixel to meaning". This paper is focused on the following questions: Who was this man? What were his scientific roots? What was his basic contribution? What are the offsprings of his work? What impetus did he give to the scientific community?. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Multidimensional multistage k-NN classifiers for handwritten digit recognition

    Page(s): 19 - 23
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (245 KB) |  | HTML iconHTML  

    This paper analyses the application of multistage classifiers based on the k-NN rule to the automatic classification of handwritten digits. The discriminating capacity of a k-NN classifier increases as the size and dimensionality of the reference pattern set (RPS) increases. This supposes a problem for k-NN classifiers in real applications: the high computational cost required. In order to accelerate the process of calculating the distance to each pattern of the RPS, some authors propose the use of condensing techniques. These methods try to reduce the size of the RPS without losing classification power. Our alternative proposal is based on hierarchical classifiers with rejection techniques and incremental learning that reduce the computational cost of the classifier. We have used 270,000 digits (160,000 digits for training and 110, 000 for the test) of the NIST Special Data Bases 19 and 3 (SD19 and SD3) as experimental data sets. The best non -hierarchical classifier achieves a hit rate of 99.50%. The hierarchical classifier achieves the same hit ratio, but with 24.5 times lower computational cost than best non-hierarchical classifier found in our experimentation and 6 times lower than Hart's Algorithm. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Handwritten address recognition with open vocabulary using character n-grams

    Page(s): 357 - 362
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (305 KB) |  | HTML iconHTML  

    In this paper a recognition system, based on tied-mixture hidden Markov models, for handwritten address words is described, which makes use of a language model that consists of backoff character n-grams. For a dictionary-based recognition system it is essential that the structure of the address (name, street, city) is known. If the single parts of the address cannot be categorized, the used vocabulary is unknown and thus unlimited. The performance of this open vocabulary recognition using n-grams is compared to the use of dictionaries of different sizes. Especially, the confidence of recognition results and the possibility of a useful post-processing are significant advantages of language models. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Feature sets evaluation for handwritten word recognition

    Page(s): 446 - 450
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (272 KB) |  | HTML iconHTML  

    This paper presents a baseline system used to evaluate feature sets for word recognition. The main goal is to determine an optimum feature set to represent the handwritten names for the months of the year in Brazilian Portuguese language. Three kinds of features are evaluated: perceptual, directional and topological. The evaluation shows that taken in isolation, the perceptual feature set produces the best results for the lexicon used. These results can be further improved combining the feature sets. The baseline system developed obtains an average recognition rate of 87%. This can be considered a good result considering that no explicit segmentation is performed. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Script and nature differentiation for Arabic and Latin text images

    Page(s): 309 - 313
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (345 KB) |  | HTML iconHTML  

    A method for Arabic and Latin text block differentiation for printed and handwritten scripts is proposed. This method is based on a morphological analysis for each script at the text block level and a geometrical analysis at the line and the connected component level. In this paper, we present a brief survey, of existing methods used for scripts differentiation as well as a general characteristics of Arabic and Latin scripts. Then, We describe our method for the differentiation of these last scripts. We finally show two experimental results on two different data sets. 400 text blocks constitute the first one and 335 text blocks compose the second. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Novel approaches to optimized self-configuration in high performance multiple-expert classifiers

    Page(s): 189 - 194
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (257 KB) |  | HTML iconHTML  

    Classifier combination and the design of multiple expert decision combination strategies are now considered to be very important issues in pattern recognition. This paper describes an investigation covering two important aspects of decision combination: optimization and generality. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • Extraction of place-name from natural scenes

    Page(s): 239 - 243
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (299 KB) |  | HTML iconHTML  

    This paper proposes an experimental character recognition system to recognise place-names written on signboards. Recently, mobile phones with digital cameras and handy digital cameras have be come popular, so we think this system is useful. In experiments, we tested a total of 112 natural scene images with 320 characters. We obtained a correct character recognition rate of 99% and a place-name recognition rate of 98% (with a rejection rate 2%). View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A learning algorithm for structured character pattern representation used in online recognition of handwritten Japanese characters

    Page(s): 163 - 168
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (366 KB) |  | HTML iconHTML  

    This paper describes a prototype learning algorithm for structured character pattern representation with common sub-patterns shared among multiple character templates for online recognition of handwritten Japanese characters. Although prototype learning algorithms have been proved useful for an unstructured set of features, they have not been presented for structured or hierarchical pattern representation. In this paper, we present cost-free parallel translation without rotation of sub-patterns that negates their location distributions and normalization that reflects feature distributions in raw patterns to the sub-pattern prototypes, and then show that a prototype learning algorithm can be applied to the structured character pattern representation with significant effect. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • A fast algorithm for finding k-nearest neighbors with non-metric dissimilarity

    Page(s): 13 - 18
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (313 KB) |  | HTML iconHTML  

    Fast nearest neighbor (NN) finding has been extensively studied. While some fast NN algorithms using metrics rely on the essential properties of metric spaces, the others using non-metric measures fail for large-size templates. However in some applications with very large size templates, the best performance is achieved by NN methods based on the dissimilarity measures resulting in a special space where computations cannot be pruned by the algorithms based-on the triangular inequality. For such NN methods, the existing fast algorithms except condensing algorithms are not applicable. In this paper, a fast hierarchical search algorithm is proposed to find k-NNs using a non-metric measure in a binary feature space. Experiments with handwritten digit recognition show that the new algorithm reduces on average dissimilarity computations by more than 90% while losing the accuracy by less than 0.1%, with a 10% increase in memory. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.
  • The effect of large training set sizes on online Japanese Kanji and English cursive recognizers

    Page(s): 36 - 40
    Save to Project icon | Request Permissions | Click to expandQuick Abstract | PDF file iconPDF (251 KB) |  | HTML iconHTML  

    Much research in handwriting recognition has focused on how to improve recognizers with constrained training set sizes. This paper presents the results of training a nearest-neighbor based online Japanese Kanji recognizer and a neural-network based online cursive English recognizer on a wide range of training set sizes, including sizes not generally available. The experiments demonstrate that increasing the amount of training data improves the accuracy, even when the recognizer's representation power is limited. View full abstract»

    Full text access may be available. Click article title to sign in or learn about subscription options.