Richard Buse

A structural and relational approach to handwritten word recognition

In this paper, we present a new off-line word recognition system that is able to recognize unconstrained handwritten words using grey-scale images. This is based on structural and relational information in the handwritten word. We use Gabor filters to extract features from the words, and then use an evidence-based approach for word classification. A solution to the Gabor filter parameter estimation problem is given, enabling the Gabor filter to be automatically tuned to the word image properties. We also developed two new methods for correcting the slope of the handwritten words. Our experiments show that the proposed method achieves good recognition rates compared to standard classification methods.

Word recognition using fuzzy logic

This paper presents an offline word-recognition system based on structural information in the unconstrained written word. Oriented features in the word are extracted with the Gabor filters. We estimate the Gabor filter parameters from the grayscale images. A two-dimensional fuzzy word classification system is developed where the spatial location and shape of the membership functions are derived from the training words. The system achieves an average recognition rate of 74% for the word being correctly classified in the top position and an average of 96% for the word being correctly classified within the top five positions.

Using Gabor filters to measure the physical parameters of lines

Abstract A new method is proposed for measuring physical parameters of lines using the response of only a bank of Gabor filters. These measurements were made without resorting to an image ruler. First we measured the length and angle of a single isolated line and then extended the method to images consisting of many lines. A constraint on this method is that the lines in the scene need to be separated and isolated by a minimum distance. Results indicated that Gabor filters can be successfully applied to the measurement of geometric properties of objects, especially where Gabor filters are already being used for processing tasks. The best accuracies in terms of measurement error for the line length and angle were 0.71 and 0%, respectively.

Feature extraction and analysis of handwritten words in grey-scale images using Gabor filters

We present a new off-line word recognition system that is able to recognise unconstrained handwritten words from their grey-scale images, and is based on structural and relational information in the handwritten word. We use Gabor filters to extract features from the words, and then use an evidence-based approach for word classification. A solution to the Gabor filter parameter estimation problem is given, enabling the Gabor filter to be automatically tuned to the word image properties. Our experiments show that the proposed method achieves reasonably high recognition rates compared to standard classification methods.<<ETX>>

Pre-processing and Feature Extraction

Many different types of features can be used to recognize handwritten words and characters. Good features should enable the system to discriminate different classes effectively, to reduce redundancy in representation and be robust to noise and deformation. In this chapter we discuss features and feature extraction techniques for handwriting recognition.

A Fuzzy Structural Approach to Handwritten Word Recognition

In this paper, we present a new off-line word recognition system that is able to recognise unconstrained handwritten words from their grey-scale images, and is based on structural information in the handwritten word. We use Gabor filters to extract oriented features from the words. A 2D fuzzy-word classification system has been developed where the spatial location and shape of the membership functions is derived from the training words. The Gabor filter parameters are estimated from the grey-scale word images enabling the Gabor filter to be automatically tuned to the word image. Our experiments show that the proposed method achieves high recognition rates compared to standard classification methods.

Markov Random Field Models for Recognizing Handwritten Words

This chapter presents a system for recognizing unconstrained handwritten words. In this system, we describe handwritten words in terms of directional line segments that are extracted by Gabor filters and we use MRF models for recognizing unconstrained handwritten words. As discussed in detail in Chapter 5, the main advantage of MRF models is that they provide a flexible and natural framework for modeling the interaction between spatially related random variables in their neighborhood systems. To deal with shape variations in handwritten words, we use fuzzy neighborhood systems and fuzzy matching measurements. The relaxation labeling algorithm is used to maximize the global compatibilities of the MRF models. We investigate the influence of neighborhood size and iteration number of relaxation labeling on recognition rates.

Markov Models with Spectral Features for Handwritten Numeral Recognition

Probabilistic models are powerful in coping with large variations in shapes, whereas Fourier spectra are suitable for describing 2-D shapes with simple closed curves. In this chapter, we introduce a hybrid recognition method that uses Markov process to model spectral features for recognizing hand-written numerals. We analyze the properties of the Fourier descriptors for spectral features derived from contours of 2-D shapes. These features can be used for 2-D pattern recognition. Section 4.1 gives a brief review of 2-D pattern recognition using the features extracted from contours. Section 4.2 introduces Fourier descriptors for spectral features. Section 4.3 presents the Markov model-based method for recognizing handwritten numerals. This chapter also presents efficient re-estimation and evaluation algorithms. The results of handwritten numeral recognition using the proposed method are given in Section 4.4.

Markov Random Field Model for Recognizing Handwritten Digits

Markov random field (MRF) models are multidimensional in nature that allows us to combine both statistical and structural information for pattern recognition. Therefore, MRF models have been widely applied to image restoration, segmentation, texture modeling and classification.

Hidden Markov Model-Based Method for Recognizing Handwritten Digits

In this chapter, we introduce the theory of HMMs and present an HMM-based method for recognizing unconstrained handwritten numerals.