In a criminal investigation, a writer identification is frequently performed to find out what kind of writer a certain letter was written. In comparing the similarity between the handwriting of both documents, it is necessary to extract a person's characteristic writing style. Although the text-dependent methods have been studied show high discrimination performance for a writer identification, the situation assumed in the actual appraisal that the same character class does not exist is not supported. In this research, we propose a writer identification method taking into account the actual appraisal situation. We define the handwriting features without dependence on character class as "personal writing style" and construct the AutoEncoder with the condition of character class to extract personal writing style from a single character sample. In the latent space trained by the conditional AutoEncoder, similar personal writing styles are mapped on to neighboring points independent of the character class. At the time of writer identification, the similarity between the handwriting feature of unknown writer and reference writer in the latent space is evaluated. In order to confirm the effectiveness of the proposed method, we conducted a writer identification experiments using ETL-1 Character Database and NIST Special Database 19 2^nd Edition. As a result, it is indicated that it is possible to extract personal writing style, which is an effective feature for writer identification, even under conditions close to the practical situation.

Musical ensemble between human musicians and computers is a challenging task. We achieve this with a concert-quality synchronization using machine learning. Our system recognizes the position in a given song from the human performance using the microphone and camera inputs, and responds in real-time with audio and visual feedback as a music ensemble. We address three crucial requirements in a musical ensemble system. First, our system interacts with human players through both audio and visual cues, the conventional modes of coordination for musicians. Second, our system synchronizes with human performances while retaining its intended musical expression. Third, our system prevents failures during a concert due to bad tracking, by displaying an internal confidence measure and allowing a backstage human operator to "intervene" if the system is unconfident. We show the feasibility of the system with several experiments, including a professional concert.

Handwriting identification is a method to identify an unknown writer by comparing a known writer's characters with an unknown writer's characters by using the homeostasis of handwriting in the characters. In handwriting identification, improving the accuracy is important in analyzing a large number of characters of the same class. However, a problem with the accuracy of handwriting identification occurs because of lack of characters of the same class. In order to solve this problem, we propose a method of extracting features of handwritten characters from a character class and generating characters of different character class based on the extracted features. It is difficult to extract handwriting features by scientific numerical analysis method, and to generate characters from extracted handwriting features. In this paper, we tried to construct a learning method using Deep Learning to generate a model of the handwriting feature extraction and the handwriting character data generative model. Using the existing neural network model, we tried to apply the generative model of the data of Japanese characters (Hiragana). Because existing models cannot capture complicated character features like Hiragana, we propose a modified method to improve character generation accuracy by adding a convolution model to the existing neural network model. Improvement of character generation accuracy by adding the convolution model was confirmed by using quantitative evaluation method of generated characters.

In order to enhance the performance of spaceborne synthetic aperture radar-ground moving target indication (SAR-GMTI) systems, multichannel systems with large and preferably nonuniform baselines are required. In this paper, SAR-GMTI algorithms for multichannel SAR systems, which we call multichannel displaced phase center antenna (DPCA), multichannel along track interferometry (ATI), and multichannel DPCA-ATI, are presented. Multichannel DPCA is a deterministic algorithm for clutter and azimuth ambiguity suppression. It successfully suppresses not only uniform azimuth ambiguities but also nonuniform isolated ones, since it does not require uniform clutter covariance assumption as adaptive algorithms do. Multichannel ATI and multichannel DPCA-ATI are the algorithms for target radial velocity estimation. Both of them reduce the target radial velocity ambiguities, which arise with the long baseline systems, by exploiting the multiple receive channel signals. And multichannel DPCA-ATI further achieves robust performance to clutter influence by suppressing the clutter and the azimuth ambiguity in advance. The performances of the proposed algorithms are shown through airborne Ku-band three-channel SAR experiments. It is shown that the multichannel DPCA suppresses strong azimuth ambiguity up to more than 20 dB, and the accuracy of the radial velocity estimation of the multichannel DPCA-ATI is on the order of 0.1 m/s. Furthermore, statistical performance analysis is presented to discuss the potential performance on the spaceborne system.

To realize 3D spatial sound rendering with a two-channel headphone, one needs head-related transfer functions (HRTFs) tailored for a specific user. However, measurement of HRTFs requires a tedious and expensive procedure. To address this, we propose a fully perceptual-based HRTF fitting method for individual users using machine learning techniques. The user only needs to answer pairwise comparisons of test signals presented by the system during calibration. This reduces the efforts necessary for the user to obtain individualized HRTFs. Technically, we present a novel adaptive variational AutoEncoder with a convolutional neural network. In the training, this AutoEncoder analyzes publicly available HRTFs dataset and identifies factors that depend on the individuality of users in a nonlinear space. In calibration, the AutoEncoder generates high-quality HRTFs fitted to a specific user by blending the factors. We validate the feasibilities of our method through several quantitative experiments and a user study.

While the speed of the Internet has been increasing, HTTP/1.1 has been plagued by head-of-line blocking, low concurrency and redundant headers. To solve these problems, HTTP/2 was standardized. This paper summarizes our experience implementing HTTP/2 in Haskell. We found several techniques to improve the performance of the header compression and identified a suitable data structure for HTTP/2 priority. Also, we showed that Haskell lightweight threads are useful for HTTP/2 where the common tactics of one lightweight thread per connection cannot be used. The HTTP/2 implementation of Warp, the popular HTTP server library in Haskell, ultimately provides better throughput than its HTTP/1.1 counterpart.

Haskell threads provide a key, lightweight concurrency abstraction to simplify the programming of important network applications such as web servers and software-defined network (SDN) controllers. The flagship Glasgow Haskell Compiler (GHC) introduces a run-time system (RTS) to achieve a high-performance multicore implementation of Haskell threads, by introducing effective components such as a multicore scheduler, a parallel garbage collector, an IO manager, and efficient multicore memory allocation. Evaluations of the GHC RTS, however, show that it does not scale well on multicore processors, leading to poor performance of many network applications that try to use lightweight Haskell threads. In this paper, we show that the GHC IO manager, which is a crucial component of the GHC RTS, is the scaling bottleneck. Through a series of experiments, we identify key data structure, scheduling, and dispatching bottlenecks of the GHC IO manager. We then design a new multicore IO manager named Mio that eliminates all these bottlenecks. Our evaluations show that the new Mio manager improves realistic web server throughput by 6.5x and reduces expected web server response time by 5.7x. We also show that with Mio, McNettle (an SDN controller written in Haskell) can scale effectively to 40+ cores, reach a throughput of over 20 million new requests per second on a single machine, and hence become the fastest of all existing SDN controllers.

This paper presents an experimental result to demonstrate azimuth ambiguity suppression with the displaced phase center antenna technique that is promising technique to realize the high resolution wide swath imagery for spaceborne synthetic aperture radar (SAR). The dual receiver antenna data was acquired by airborne Ku-band SAR. In this paper, we propose a signal compensation procedure to cope with beam squint for the following coherent signal processing such as the dual channel reconstruction algorithm. The azimuth ambiguity suppression capability is evaluated by comparing the image reconstructed from dual-receiver channel to the single receiver channel image. The experimental result shows significant azimuth ambiguity suppression.

Passive bistatic radars (PBRs) exploit existing transmitters such as TV broadcasts, radio broadcasts, Global Navigation Satellite System (GNSS), etc., as the source of illumination. PBR consists of two receivers, with one antenna pointed at the source and the other at the target, and the target range is determined by correlating the signal scattered by the target with the signal directly arrived at the receiver. Since PBR dode not transmit any wave form, it consumes lower power, and no frequency allocation is required. We are investigating the target imaging capability of PBR. In This paper an experimental study on obtaining In verse Synthetic Aperture Radar (ISAR) image with cross range resolution of a few meters, using terrestrial digital TV based PBR. In this study, we conducted a field experiment to observe some aircraft targets. Due to the relatively low frequency band employed by the digital TV broadcast system, a very wide synthetic aperture angle is required to obtain a few meters cross-range resolution. Using the data acquired in the experiment, we show the limit of the range Doppler processing approach and we investigate the advantage of the Polar Format Algorithm (PFA). We also discuss the issue on autofocus and show results of initial study on autofocus algorithms.

Abstract A weight-balanced tree (WBT) is a binary search tree, whose balance is based on the sizes of the subtrees in each node. Although purely functional implementations on a variant WBT algorithm are widely used in functional programming languages, many existing implementations do not maintain balance after deletion in some cases. The difficulty lies in choosing a valid pair of rotation parameters: one for standard balance and the other for choosing single or double rotation. This paper identifies the exact valid range of the rotation parameters for insertion and deletion in the original WBT algorithm where one and only one integer solution exists. Soundness of the range is proved using a proof assistant Coq. Completeness is proved using effective algorithms generating counterexample trees. For two specific parameter pairs, we also proved in Coq that set operations also maintain balance. Since the difference between the original WBT and the variant WBT is small, it is easy to change the existing buggy implementations based on the variant WBT to the certified original WBT with a rational solution.

In this research, we paid attention to calligraphy education. In current calligraphy education in Japan, teachers evaluate character written by students and they teaches based on the written character. Professional in calligraphy can estimate writing process and balance of character which are important points for evaluation from written character by estimating movement of contact shape (contact face with a paper and a brush). But it takes a lot of time for students to learn correct writing process of character and balance of character in this education way. If teachers and students can know a movement of the contact shape, the calligraphy education will be more efficient. In case of taking an image of a brush by using a general camera setting, it is difficult to separate a character written by a black ink and a black brush to detect contact shape because both are black and the contact shap is hided under the brush. Therefore, we propose new camera setting system as Horizon View Camera (HVC) that is a special camera setting, and contact shapes can be detected and reconstructed by using this camera system. And, we evaluate balance of white space which means unwritten space on a paper for evaluation of a character. Correct white space in calligraphy is not just unwritten space but is influenced by movement of contact shape. Thenby using HVC system, movement of contact shape which is difficult for amateur to estimate from written character can be detected dynamically and evaluation of white space in considering of influence of contact shape movement can be improved.

In this paper we propose a new effective method for low-resolution character recognition from motion images. This method extracts the feature directly from low-resolution images first, and reconstructs the high-resolution feature. Since low-resolution characters of motion images are recognized, a general idea is to reconstruct a high-resolution image first by using these low-resolution images, and it is extracted the feature from that high-resolution image. Feature extraction process of our proposed method has different view of the general idea. Our proposed method has higher accuracy than the general idea in low-resolution characters recognition. We show advantages of it by comparative recognition experiments under the database images and motion images.

The main purpose of our research is to evaluate which subject's facial parts are most effective at making the difference between men and women. We prepared the four directional feature fields on multiple facial parts images, such as face, jaw, lip, nose, eyes, and R eye. Then, we recognized these images by using the linear discriminate analysis. Furthermore, we analyzed the feature space by using within-class variance between-class variance ratio. After experimenting on a large number of subjects, we conclude that face and jaw make good feature spaces. Moreover, we discovered that recognition rate changes with ages. Accordingly, a possibility that each man and woman had the age, which is easy to distinguish gender, was suggested.

In this paper, an approach to detection of caption text invideo frames is described. Text recognition in video can beapplied to various applications, however there are stillproblematic issues such as insufficient resolution,complexity of layouts and backgrounds. This study attemptsto solve these problems with a segmentation-free approach,called MAP matching method. Besides extending the methodto grayscale images, a strategy for character size variationusing Gaussian filtering and multi-sized reference patternsis discussed, as well as a method for detecting framescontaining caption text. Results show the proposed matchingmethod is able to detect characters of unknown size incaption text. Although over-detection is not negligible,verifying the positions of detected characters can identifythe location of keywords with practical precision. It is alsoshown that the frames containing caption text are detectedwith nearly 98% accuracy.

In this paper, we propose a novel method to estimate the slant of omnidirectional image sensors which can acquire images of an environment with a FOV close to 360/spl deg/ /spl times/ 180/spl deg/. The proposed method consists of two steps: a low resolution voting followed by a least squares method (LSM). First, the direction of each edge pixel in the omnidirectional image is computed and projected to the plane of Z = 1. The biggest peak formed by the vertical edges, which are prevalent in indoor and urban scenes, is detected. Then, edge directions whose projections to the voting plane pass near the peak are used to estimate the slant of the sensor by LSM. Experimental results in a real environment show the effectiveness of the proposed method.

This paper describes a novel algorithm for robust object detection and segmentation, which is based on a new robust dissimilarity measure called as radial reach correlation (RRC). The capability of detecting moving objects from a complex background is one of the most fundamental technology for intelligent surveillance systems. The RRC is a new robust dissimilarity measure and has a well-formed probabilistic model of binary or normal density. The RRC evaluates the local texture between a background image and the current scene and realize robust object detection under poor conditions. To demonstrate the effectiveness of our approach, we present experimental results from real world all-directional images provided by the stereo omni-directional system (SOS).

For marketing research, it is important to know what customers are interested in, and to evaluate the effectiveness of advertisements. Nowadays, marketing information regarding the evaluation of advertisements is mainly obtained from researchers' activities, which are difficult and time consuming. The research activities can be improved by use of surveillance systems with computer vision technologies. In this paper, a marketing information system that detects customers' attention behavior with computer vision is proposed. Cameras are attached to the display that presents advertisements in the system. Thus, the proposed method mainly consists of frontal face detection with "Four-Directional Features (FDF)" and calculating the spatial position of a face by stereovision. It is supposed that this system enables long time surveillance and uniform evaluation of presented advertisements. In addition, it is easy to distinguish which advertisements attract the most attention. Experimental results show that this system can detect attention behavior.

For most of the existing pointing systems proposed in the past, the pointing target was limited to a specific surface in the space, usually an object on a 2-D screen. This paper presents a new solution for estimating the pointing target inside a 3-D space, such as a room. A special set of stereo cameras acquires images and depth maps from the entire environment. Using a background subtraction method, our system detects the foreground pixels, analyzes them and extracts the user's face and hand regions. Considering that the pointing direction is the straight line connecting the face and the hand, the pointing target can be seen as the intersection of that line with the 3-D model of the environment, allowing the calculation and the display of the person's viewpoint.

In this research, we introduce a reasonable noise model for range data which is obtained by a laser radar range finder, and derive two simple approximate solutions of optimal local plane fitting the range data under the noise model. We compare our methods with general least-squares based methods, such as Z-function fitting, the eigenvalue method, the maximum likelihood estimation method, and the renormalization method, an iterative method to obtain the optimal fitting of planes of range data under the noise model. All the methods are compared and evaluated using both synthetic range data and real range data with ground truth. From the experimental evaluation results, the proposed methods are shown to be effective, and the general least-squares-based methods are shown to be unsuitable for the assumed noise model.

This article discusses face and hand sign recognition and its applications. We have already proposed a method that can recognize faces and hand signs using hierarchical discriminant analysis. Here, we implement this method on a laptop computer and develop a real-time system that can control home appliances by recognizing faces and hand signs. Our system can control various appliances in the home by remote control. Moreover, since our system can recognize the user, it can assign distinct commands to the hand signs of each user and limit the appliances and functions that users can control.

This paper describes an identification method for book-cover images. It is used in a novel library system using images of books and the borrowers' faces. A returned book is identified from the lent books by processing the images of the book covers. A low-resolution field with four directional features is used to identify the books. These features can be compatible with computation time and performance. To identify a series of books which are similar, but different in just a small area, partial area matching is used. Experiments show the effectiveness of the proposed methods.

We propose a multi-camera system that can track multiple human faces and hands as well as focus on face and hand gestures for recognition. Our current system consists of four cameras. Two fixed cameras are used as a stereo system to estimate face and hand positions. The stereo camera detects faces and hands by a standard skin color method we propose. The distances of the targets are then estimated. Next to track multiple targets, we estimate the positions and sizes of targets between consecutive frames. The other two cameras perform tracking of such targets as faces and hands. If a target is not the appropriate size for recognition, the tracking cameras acquire its zoomed image. Since our system has two tracking cameras, it can track two targets at the same time. To recognize faces and hand gestures, we propose four directional features by using linear discriminant analysis. Using our system, we experimented on human position estimation, multiple face tracking, and face and hand gesture recognition. These experiments showed that our system could estimate human position with the stereo camera and track multiple targets by using target positions and sizes even if the persons overlapped with each other. In addition, our system could recognize faces and hand gestures by using the four directional features.

We propose a multi-camera system that can focus on faces and hand gestures for recognition. The system consists of four cameras. Two fixed cameras are used as a stereo system to estimate face and hand (F&H) positions. F&H regions are detected by a standard skin color method. The distances of the targets are then estimated. The positions and sizes of targets are evaluated for tracking between consecutive frames. The other two cameras perform tracking of such targets as faces and hands. If a target is not the appropriate size for recognition, the tracking cameras acquire its zoomed image. We propose a hierarchy linear discriminant analysis that distinguishes between faces and hands on the first step, and then recognizes face and hand gestures in each discriminant space by using four directional features. The experimental results show that our system could distinguish between faces and hands, as well as recognize face and hand gestures by using the four directional features.

For on-line character recognition, we have designed clockwise/counterclockwise direction-change features, replacing the eight kinds of direction-change features of the previous method. These new features express where and in which direction (clockwise/counterclockwise) each of the character's coordinates change. These features use only two kinds of features patterns. In this paper we propose a new method using both directional features and clockwise/counterclockwise direction-change features. The results of experiments with freely written Japanese characters demonstrate that our new method reduces the size of the standard feature patterns and shortens the processing time over those of our previous method while maintaining a high recognition rate.

We propose a new online recognition method to recognize handwritten cursive-style Japanese characters correctly. Our method simultaneously uses both directional features, otherwise known as offline features, and direction-change features which we designed as online features. The direction-change features express where in the mesh and in which direction the character's coordinates change. These features express both written strokes in the pen-down state and unwritten imaginary strokes in the pen-up state. The recognition rate was improved by our method over the traditional method using only directional features.

We found suitable direction-change features of the imaginary strokes in the pen-up state for online handwritten cursive character recognition. Our method simultaneously uses both directional features, otherwise known as off-line features, and direction-change features, which we designed as on-line features. Direction-change features express both written strokes in the pen-down state and unwritten imaginary strokes in the pen-up state. It is important to get suitable direction-change features when using this method. We tried to examine the influence on character recognition rates when changing the functions used to get each direction-change feature based on the imaginary stroke lengths. Then, we found that the best function is the function which puts no weight on the imaginary stroke lengths. The recognition rate for freely-written Japanese characters was improved from 82.37% to 86.32% by our new method using the best function as opposed to our old method using a function which gets each direction change feature in inverse proportion to the imaginary stroke lengths.

Some of the authors previously proposed a few nonlinear normalization methods using line density for handwritten characters and it was verified that those nonlinear methods were more effective than a linear method. Those nonlinear methods, however, could not use two-dimensional information of the line density, because the line density was projected onto the axes of coordinates. The paper proposes an extended nonlinear normalization method which uses two-dimensional information of the line density. It is confirmed that the proposed method is more powerful than conventional methods by experiments tested on the handwritten character database ETL-8.

We propose a new on-line recognition method to recognize handwritten cursive-style characters correctly. Our method simultaneously uses both directional features, otherwise known as off-line features, and direction-change features, which we designed as on-line features. These features are expressed in the divided meshes of the character area. The directional features express the directions between character's coordinates within the meshes. The direction-change features express where in the mesh and in which direction each direction of the character's coordinates change, and express where the circular parts of the character are in the mesh. These features express both written strokes in the pen-down state and unwritten imaginary strokes in the pen-up state. Our method recognizes an inputted character by comparing the inputted character's features and the features of standard characters. The recognition rate was improved by our method with directional features and direction-change features as opposed to the traditional method with only directional features. Moreover, the recognition rate was also improved by considering imaginary strokes in the pen-up state.

The most important problem on automatic seal impression verification is to absorb the various impression quality depending on the various affixing conditions. Over the past years, many studies have been made on absorbing the various impression quality using a two-dimensional image as reference and input. But it is difficult to verify each of a large number of input patterns by a sole two-dimensional reference pattern. This paper proposes a seal identification method using a three-dimensional image (range image) as reference. As results of the verification with actual seal impressions, the verification rate is improved from that by using two-dimensional reference impressions.

Probabilistic relaxation labeling processes have been widely used in many different fields including image processing, pattern recognition, and artificial intelligence. However it is impossible to express conviction degrees to be related to human subjectivity. We propose a novel relaxation scheme called "robust relaxation" for structural matching under uncertainty. Since this method is constructed on the Dempster-Shafer measure, the above problem can be solved.

We propose an on-line handwriting character string separation method which uniformly deals with the character separation features. We devised the string separation method using a multi-level network expression and a method using a unified network expression. The separation performance was improved by unifying the physical feature and logical feature using the network expression. Furthermore, we could speed up the string separation process with the unified network expression.

Internet text mail has been developing to satisfy various user requests, such as transporting non-textual objects and privacy enhancements. While MIME redefined the mail body format to support non-textual objects and multipart structure, PGP provides encryption and digital signature features for text mail. MIME however does not provide privacy services whereas non-textual objects cannot be exchanged with PGP. It is of recent interest to integrate PGP and MIME so that users can make use of these two services at the same time. This paper describes an integration of PGP and MIME. Our scheme embeds PGP objects into MIME and maintains backward compatibility with PGP. It is possible to encrypt, sign, and sign-then-encrypt non-textual objects, single-parts in a multi-part, an entire multipart, etc. We also explain our viewing and composing mechanisms that allow users to handle PGP/MIME messages intuitively without format restrictions.

Classical probabilistic relaxation method has been widely used for solving optimisation problems in various fields, including image processing and pattern recognition. However we realize that there exist cases in which a probability theoretic model is not adequate, especially there exists incompleteness in available information by noises. In order to solve the problem, this paper proposes a relaxation matching method based on Dempster-Shafer theory. Then the update process in probabilistic relaxation method is derived as a special case of Dempster's combination rule (1967) in DS theory.

In order to develop a shape and decorative pattern database of old pottery and ceramics objects, it is necessary to obtain the surface pattern on the 3-dimensional shape of the object. This paper describes the recovery of the revolution surface from a monocular image of unknown camera parameters and the retrieval of the 2-dimensional pattern. The camera parameters are obtained by using a genetic algorithm (GA). After the revolution surfaces are reconstructed, these surfaces are developed into a 2-dimensional plane. We show that a scanner-digitized image of an old ceramic object can be analyzed by GA to reconstruct the revolution surface and to develop the 2-dimensional image on the 3-dimensional object.

This paper describes several new image understanding methods based on parallel operation. There are several constraint satisfaction approaches using a energy minimization. We show how we reconstruct three dimensional surfaces from contours without elevation data and sparse points of known elevation data by using this approach. We also introduce Active Net using this approach and apply this model to segmentation and binocular stereo matching. We experimented with these energy minimization approaches to solve the problems of early and intermediate level of computer vision and show some of the results of our recent research. 1

This paper presents a method to obtain an image processing procedure by showing an original image and a target image. There are some preceding works concerns with automatic acquisition of image processing procedures by combining subroutine packages. In our approach, we only use the hierarchical mathematical morphology procedures, i.e., erosion and dilation operations. Since the subroutine packages are macro procedures, we use micro operations for constructing more flexible visual procedures. We use genetic algorithms (GA) to search procedures. Utilizing a parallel searching function of GA and introducing a concept of era, we construct hierarchical procedures of mathematical morphology.

We have proposed the MAP (Multi-Angled Parallel) operation method which extracts geometric features from a topographic map using erosion-dilation operations on a directional feature field. We have also proposed the MAP matching method which recognizes fixed-shaped symbols in the topographic map by using parallel operations on the same directional feature field. We have shown that all kinds of feature extraction and recognition can be effectively performed within the framework of parallel operation. In this paper, we show that elevation values in a topographic map can be recognized by a combination of MAP operation with the MAP matching. Numerals and symbols are recognized by the MAP matching method, while small dots and lines are extracted by the MAP operation method. These results are used together to determine the value, position and attribute of elevations in topographic maps.

A training method of dictionary characters is presented for handwritten character recognition by the model based on relaxation approach. The dictionary characters are constructed by analyzing the structure of several training characters using curved line approximation and relaxation matching. The performance of the constructed dictionary characters was verified by the experiment. 956 (chars.) /spl times/ 20 (sets) in the standard database ETL-8 (Japanese KANJI and HIRAGANA) are used as sample characters for training and 10 (sets) are used as test characters (5 known sets and 5 unknown sets). The mean of the number of constructed dictionary characters is 3.81 for each category, and recognition rate is 99.96% on the known sets and 99.73% on the unknown sets.>

A new method for stamped character recognition is proposed. Conventional techniques have used a camera to observe the characters and get the image as a gray-level. However, it has been difficult to get an exact image, because gray-level noise occurs at some distortion around the characters, which makes a shadow on the characters, and also at these parts of stamped surface which are dirty or painted. The approach presented can solve these problems by using a range finder system as the observation system. The method observes the stamped characters as range image. The authors compare the characteristics of a range image to the characteristics of a gray-level image, especially about the noise, and show a noise reduction method for range images. Also, the performance is verified by an experiment.>

Various combinatorial optimization methods have been considered independently. The purpose of this study is to organize theoretically those optimization methods, which differ by their natures, and to indicate clearly what are common and what are different among those methods. The authors deal with a labeling problem in the optimization methods and propose a generalized algorithm to solve the problem. They show that the algorithm can derive the conventional optimization methods, such as the relaxation matching method, neural networks, and the dynamic programming matching method.>

The authors reconstruct three dimensional surfaces described by contours on topographic maps. When the elevations of the contours are available, this is an easy problem to solve. However, the authors do not have these elevations and can use only sparsely distributed elevation values noted on the map. This is a difficult problem even for humans to solve. They have developed a method for reconstructing three dimensional surfaces using only contours without matching elevations and these sparsely noted elevation values on the map as input. Numerals and symbols are recognized by the multitangled parallelism (MAP) matching method, while small dots and lines are extracted by the MAP operation method. These results are then used to used to determine the value, position, and attributes of the elevations marked on the topographic maps. The authors define an energy function for reconstruction of the surface. The function integrates three kinds of constraints: smoothness, fitness, and contour. The energy function can be minimized by solving a large linear system of simultaneous equations. The experimental result is a surface reconstructed from a 25,000:1 scale topographic map of the Tsukuba area.>

In order to verify effectiveness of the perturbation method in the optical character recognition procedure, the improvement of performance was experimentally confirmed by applying perturbation to all six parameters of two-dimensional affine transformation. The technique of character recognition consisted of reciprocal directional features, directional features, and correlation, and was applied to the character database, the handwritten Kanji database ETL8b2. As a result of the experiment, the recognition accuracy was improved by 5%, and the maximum correct recognition rate was 97.4%. The problem to be solved is the enormous amount of computation of the procedure requires. There are a large number of parameters to which the perturbation is applicable other than affine transformation, and extension of this approach makes the problem diverge.>

One of the most difficult and important problems encountered in the automatic digitizing of graphical topographic maps is the identification and separate digitizing of different kinds of features. Essentially, in topographic maps, there are two kinds of geometric features: linear features, such as roads and railways that have an arbitrary length, and symbols, which indicate a type of building or area of land usage or even numerical information. These two types of features are extracted and recognized by using methods based on multiangled parallelism (MAP). The MAP operation method performs parallel calculation on directional feature planes. The linear features are extracted using erosion-dilation operations on the directional feature planes, and the symbols are extracted using a reformalized and parallel version of the generalized Hough transformation on the same directional planes, which is called the MAP matching method. The methods have been applied to a 1/25000 scale map.>

Describes a method for tracking a deformable object in a time-varying image sequence and analyzing the motion of the object. The approach is based upon modeling the boundary of the object as a new active contour model defined by an energy functional in terms of intra- and inter-frame constraints, and then finding the object boundary in the temporal images by minimizing the energy functional using dynamic programming in multiple scales. Motion analysis is carried out by analyzing some characteristic points which are extracted as local curvature extrema along the contour. The proposed method has been successfully applied to the analysis of a slug (Limax flavus) in motion with smooth deformation, in the study of the mechanisms of learning and memory formation in slugs.>

As a recognition method of characters stamped on the metal, a process without binarization is proposed. Segmentation and recognition are performed simultaneously by exhaustive pattern matching at every location using multi-angled parallel matching method, where multiple directional planes and graylevel feature are combined as an input feature. The effectiveness is proved by experiments for laser-marked alphanumerics.>

Describes a method for recovering volumetric shape descriptions of 3D objects from range data. The models of shape primitives are superquadrics with a global deformation of tapering. The authors first define a new error-of-fit measure for fitting superquadrics to range data. The measure evaluates the difference in both depth and surface orientation between input 3D data and the model surfaces. Simulated annealing is then employed for finding a global minimum of the error-of-fit over all possible values of model parameters including translation, rotation and tapering.>

Research and development of OCR systems are considered from a historical point of view. The historical development of commercial systems is included. Both template matching and structure analysis approaches to R&D are considered. It is noted that the two approaches are coming closer and tending to merge. Commercial products are divided into three generations, for each of which some representative OCR systems are chosen and described in some detail. Some comments are made on recent techniques applied to OCR, such as expert systems and neural networks, and some open problems are indicated. The authors' views and hopes regarding future trends are presented.>

The multiangled parallelism (MAP) concept, which defines directional operations with directional elements, is discussed. To achieve the MAP operation, a parallel architecture in which MAP operations are summarized as single or cyclic, unary or binary, logical operations is proposed. The hardware design is simulated using an existing image processor. It is demonstrated that geometric figures can be extracted from a topographical map by first extracting the core features and then expanding it gradually.>

Research work dealing with some of the problems in computer vision (CV) and image understanding (IU) is described. Attention is given to progress in CV and IU involving active range finders, passive stereo sensing, 3D reconstruction, 3D scene analysis, dynamic scene analysis, automatic knowledge acquisition for model- and strategy-based systems, and autonomous vision systems.>

A handprinted Chinese character database is introduced. Features of characters in the database are analyzed in order to develop a handprinted Chinese character recognition method based on relaxation matching. An analysis shows that this method, which is one of the most accurate methods for handprinted Kanji, is also successful in recognizing handprinted Chinese characters. The results will be used to modify the method to improve its performance for handprinted characters.>

A model-based dynamic programming matching method is applied to extract the optimally fitting shape from each static frame in a sequence of echocardiograms, while allowing for a certain amount of change of length of each side of a model polygon. Discussions and experiments are presented in order, from general to specific. This is managed by beginning with the method for static 2D images, and adding integration and differentiation along the time axis. A temporal continuity condition is also considered. Finally, the knowledge that the movement is periodic and the moving region is restrictive is used to constrain the shape and position in the matching process.>

MIRACLE-IV is discussed, which is capable of obtaining an internal structure of an object from a series of silhouette images with no initial explicit models about the object. The images are derived from only one object, but the form is varied. The system is composed of two subsystems: a model acquisition part (the modeler) and an image processing strategy part (the strategist). On the assumption that the object consists of hinges, slides, and solids, the modeler learns their number in the object and the relationship among them. The strategist binds the functional features as hinges or slides with visual features in the actual image data. The image processing sequence for the extraction of the visual feature is not given but is learned automatically through trial and error.>

This paper proposes a method of extracting features of handwritten Japanese characters by using spline approximation. This method represents a character by contours expressed by well-approximating functions and steady breakpoints which characterize the connection of the strokes so that it provides proper features for handwritten Japanese character recognition with relaxation matching. Its effectiveness tested on a Japanese character database is also shown.

Relaxation method attracts attention as an effective method in image processing and pattern recognition. From the first, the relaxation operation has been said to be capable of parallel processing. But previous hardware implementations were specialized to each application. This paper presents a parallel architecture of a relaxation processor which can be applied to general use.

Machine recognition of handprinted Chinese characters has recently become very active in Japan. Both from the practical and the academic point of view, very encouraging results are reported. The work is described systematically and analyzed in terms of so-called feature matching, which is likely to be the mainstream of the research and development of machine recognition of handprinted Chinese characters. A database, ETL8 (881 Kanji, 71 hirakana, and 160 variations for each category), is explained, on which many experiments were performed. Recognition rates reported using this database can be compared, and so somewhat qualitative evaluation of these methods is described. Based on the comparative study, the merits and demerits of both feature and structural matching are discussed and some future directions are mentioned.