/ Shunji Mori / Professor
/ Daming Wei / Visiting Professor
/ Jintae Lee / Associate Professor
/ Tony Y. T. Chan / Assistant Professor
/ Yu Nakajima / Research Associate
First and foremost, the Image Processing Laboratory engages in research and development of image pattern recognition systems. More specifically, as can be seen from the background and the recent research publications of the members of the laboratory, character recognition was our current focus. However, Prof. Lee joined to our laboratory last year and doing research actively who is an expert of model-based synthesis and recognition. Therefore, a new dimension was added to our laboratory. Furthermore Dr. Wei joined to us as a visiting researcher who is famous for his heart model and known worldwide in the field of Medical Engineering.
Related to the recent involvement of multimedia systems, character recognition has been noted by many researchers and engineers. On the other hand, character recognition techniques are generally divided into off-line and on-line methods. Historically speaking, these techniques have been developed separately. However, both techniques can be developed together in principle. The common approach makes possible recognition flexibility in such a way that the usual constraints being imposed on the on-line techniques can be removed. For example, writing order and number of stokes constituting a character are typical of such constraints. In this connection in SCCP project the students has developed so-called Kanji-learning System which evaluate written character and output some score with some instruction such as writing order and balance of the shape of the written character. The system was demonstrated in many exhibitions.
The contribution to the domestic industry is very important mission of our University. In this connection we started the development of a specific document reading system with a contract with a company. This project is going well. In this connection, Prof. Mori and Mr. Nakajima have engineering background. Actually, Prof. Mori was a leader of an OCR group when he worked for Ricoh Research and Development Center and developed a very powerful OCR package which is used in practice. One of the members of the OCR group was Mr. Nakajima, who has joined us as a research associate. He is well-acquainted with different computer systems and also works as an instructor in programming. The OCR group were awarded the Excellence in Programming Prize from the Ministry of Post and Telecommunications. Associate professor J. Lee has broad research background, ranging from computer graphics and computer vision to visual language and visual communication. While diverse in content, his research nonetheless focuses on the basic underlying theme: how do human beings understand the complicated, static and dynamic images and what are the computational means and techniques we can employ to assist people in creating, conveying, or understanding these images. His early research was devoted to model-based object recognition, and showed that both realistic animation and efficient recognition of the human hand are possible by employing well-constrained hand models. His recent research has been developing virtual brushes useful for oriental painting and on-line character recognition. With the recent advent of Internet, he plans to implement these simulations and new visualizations within virtual worlds on the Internet.
Assistant Professor Chan is investigating unifying pattern recognition and artificial intelligence approaches based on models proposed by Prof. Lev Goldfarb of the University of New Brunswick and himself.
A visiting researcher and Professor Wei did his research very actively. That is, he published three scientific papers in referred journals. In addition, he presented 4 papers in referred conferences (two international conferences and two Japanese conferences). As a memorial lecture in Aizu-wakamatsu High-tech conference , he was invited to present a two-hour lecture entitled "Biomedical Engineering and Medical Industry".
Refereed Journal Papers
Two methods for stroke segmentation from global point of view are presented comparatively. One is based on thinning method. The other is based on contour curve fitting. For both cases an input image is binarized. For the former, Hilditch's method is used. Then crossing points are sought, around which some domain is constructed. Outside the domain, a set of line segments are identified. These lines are connected and approximated by cubic B-spline curves. Smoothly connected lines are selected as segmented curves. This method works well for some limited class of crossing lines, which are experimentary shown. The other is that a contour line is approximated by cubic B-spline curve, along which curvature is measured. According to the extreme points of the curvature graph, the contour line is segmented, based on which the line segment is obtained. Experimental results are shown for some difficult cases.
Hypertrophic cardiomyopathy (HCM) was simulated with a computer heart model having realistic shape and rotating fiber orientation in order to elucidate possible mechanisms for abnormal ECG findings. The disarray of myocardial muscle in HCM was simulated by assigning random fiber direction and isotropic electrophysiologic properties to abnormal hypertrophic regions, in contrast to the anisotropic modeling for normal myocardium. With these models, main ECG features including abnormal Q wave and QS pattern were reproduced, which are comparable with clinical findings. This study suggests that the change in anisotropy in the hypertrophic myocardium is likely to be a main factor responsible to the ECG features of HCM.
This paper proposes a low distortion filter method for rejection of muscle noise in multi-lead ECG systems. This approach combines low-pass filtering with a modified version of source consistency filtering. The low-pass filtering splits the raw ECG signal into a muscle-noise-free part and a muscle-noise-overlapping part. The modified source consistency filtering then extracts the signal components from the muscle-noise-overlapping part. The extracted signal components are restored to the muscle-noise-free part as the output. The performance of our method was verified with simulated and clinically recorded ECG signals. The simulated ECG signals were created from computer simulation using 3-dimensional, realistically shaped heart and torso models. The ideal signals are superimposed with white noise to simulate muscle noise, and with a low frequency sine wave to simulate baseline drift. In verification, our method was compared with Butterworth low-pass filters. The results show that our method can effectively reduce muscle noise with less distortion of the QRS wave than conventional low-pass filters.
An Electrocardiographic computer simulation was conduced to study the feasibility of predicting accessory pathway locations in Wolff-Parkinson-White (WPW) syndrome with body surface potential Laplacian maps. Three-dimensional, realistically-shaped heart and torso models were used. Ten accessory pathways (APs) around the atrioventricular ring corresponding to Gallagher et al were set in the heart model, and body surface Lapacian and potential maps of WPW syndrome with single or multiple APs were simulated and compared to each other. In simulations with a single AP in the anterior walls, the maximum-minimum pairs in Laplacian maps appeared similar to that in potential maps in their locations and orientations, but the maiximum-minimum pairs in Laplacian maps were shaper and more localized than in potential maps. In simulations with a posterior AP or multiple APs, the maximum-minimum pairs in the Laplacian maps showed feathers correlative to the AP locations, but no such features were found in potential maps. These results suggest a possibility of Laplacian maps to be used, as a non-invasive method, in predicting accessory pathways locations in WPW syndrome.
Soft brushes having bristles that elastically respond to the force exerted by an artist against the paper are modeled and effectively used to simulate oriental black ink painting.
Sign language is the primary communication means for the hearing impaired because typewriting or handwriting cannot provide equivalent power of expression, naturalness, or convenience. The capability to computer-process sign language has consequently received much research attention. This paper presents the 3D hand model for performing hand sign recognition based on 3D structural features of the hand, that is, flexion of joints and contactness of fingers.
One of the important questions in visual language research is how to specify a visual language. Up to date, modified phrase structure grammars were used for specification of visual languages and the focus was primarily on stationary two-dimensional pictures. In this paper we propose Generalized Logic Grammar as a general mechanism for describing the syntax of high dimensional visual languages. We compare the notion with other existing methods, and give a number of examples including dynamic visual languages that illustrate its strength.
Every APLer knows the trick that one way to cut down on the CPU time is to avoid writing explicit APL loops. But it is not so common knowledge that when one has a choice between a sequential and a parallel algorithm for a given problem, implementing the parallel algorithm using an APL2-like language can achieve a substantial saving in CPU time---even on a sequential machine. This article will demonstrate the truth of that by a case study and hopes to persuade the readers to find parallel algorithm solutions to problems before implementing them in APL.
A general (non-heuristic) computational analytical model to tackle the difficult unsupervised inductive learning problem is proposed by making some additions and modifications to an existing metric model so that the model is more elegant and able to handle the unsupervised case. It turns out that it is instructive to treat, in essence, the supervised problem with noise as an unsupervised problem. We demonstrate the success of the new model on the benchmark XOR (exclusive-or) and parity problems by showing how the inductive agent successfully learns the weights in a dynamic manner that would allow it to distinguish between bit-strings of any length and unknown labels.
Refereed Proceeding Papers
The disarray of hypertrophic myocardial tissues were modeled to possess random fiber directions in a 3-D whole-heart model with rotating anisotropy. ECGs simulated with this model showed abnoraml Q waves that are found in clinical practice. A possible mechanism responsable to this heart abnormability was suggested.
Microscopic identification is currently the major method for authenticity examination of traditional Chinese patent medicine (TCPM). With this method, preparations of powdered TCPM drug are observed under photo or polarizing microscopes and mircrographs are compared with those made from standard crude drug samples. This procedure is now performed manually. Since manual identification is time exhausting, it cannot meet the increasing needs of TCPM market. In addition, microscopic identification needs specialists who must be very familiar with the microscopic features of several hundreds of basic crude drugs and are hardly found in both China and Japan. This paper gives preliminary results to show feasibility of computerizing the microscopic identification procedure using image features.
We have developed an electrocardiographic tool built in a memory card of a electronic mobile notebook (Zaurus, Sharp Corp.). The purpose of this development is to provide a mobile tool for medical and paramedical persons who can "do ECG" anywhere, anytime. A touch-pen is the only thing needed for operation. The tool is designed mainly as an ECG manual for reading and interpreting ECGs. Typical ECG waveforms and key points of interpretation for main types of heart disease can be searched. Principles of medication dealing with arrhythmias are incorporated based on clinical experience of one of the authors. In addition, many other functions and information are included - from basic ECG knowledge to a concise Japanese-English dictionary of ECG terminology.
When approximate electroencephalographic field of brain with a two-dipole model, the equivalent dipoles can be calculated. Since values of dipolarity, or goodness-of-fit, is time varying, a synthetic index of dipolarity, that features best fitting and specify each individual, is necessary in application. Certainly, this index should be extracted automatically to ensure accuracy. In this study, a computer program is developed to extract an dipolarity index and the reliability of this index is examined with clinically recorded ECG data.
This paper presents a feature-based framework for sign gesture representation. Sign languages are being increasingly accepted as languages with their own grammatical structure, and many attempts are made to process sign language by computer. However, systems built without sound representational model of sign gestures are apt to have serious limitations. We present a new representational framework of sign gestures that accounts for morphological or phonological structure of the sign gesture and has semantics which allows easy computer implementation of sign language.
A model is proposed for learning to classify patterns under the Euclidean setting. Each pattern is represented by a vector in a fixed D-dimensional Euclidean space. Patterns are divided into training and test sets. Eleven experiments were performed. The proposed naive learner is found to be extremely fast and yet the correct classification rates are respectable even when compared with some of the best known rates.
Chapters in Books
Two methods for stroke segmentation from global point of view are pr esented comparatively. One is based on thinning method. The other is based on contour curve fitting. For both cases an input image is binarized. For the former, Hilditch's method is used. Then crossing points are sought, around which some domain is constructed. Outside the domain, a set of line segments are identified. These lines are connected and approximated by cubic B-spline curves. Smoothly connected lines are selected as segmented curves. This method works well for some limited class of crossing lines, which are experimentary shown. The other is that a contour line is approximated by cubic B-spline curve, along which curvature is measured. According to the extreme points of the curvature graph, the contour line is segmented, based on which the line segment is obtained. Experimental results are shown for some difficult cases.
Academic Activities