In the practical application of machine learning, the opaqueness of models often poses significant challenges. While decision trees are known for balancing representability with interpretability, enabling humans to understand decision rules, their interpretability decreases as the complexity of the task increases and the tree size expands, making it difficult to trace and interpret the decision flow. In this paper, we introduce a new variant of decision tree called Bivariate Nonlinear Node Decision Tree (BINN-DT), designed to enhance the interpretability of decision trees. BINN-DT selects bivariate features at each node and utilizes nonlinear splitters to learn the data splitting rules. Additionally, each node visualizes the relationship between the data distribution and split boundaries through a two-dimensional map using the selected bivariate features. Our experiments compared the proposed BINN-DT method with traditional univariate decision trees. The results demonstrate that our approach not only maintains classification accuracy but also produces more compact models. BINN-DT clearly depicts the entire decision boundaries of a model as a tree-structured collection of two-dimensional maps with the bivariate feature axes selected from the entire features. Our method significantly improves the interpretability of models by producing more compact models than the traditional decision trees, without sacrifice of accuracy.

In recent years, with the increasing demand for edge AI, the need for faster AI inference on embedded CPUs is growing. As a result, researchers have focused on various redundancies in neural networks and proposed network acceleration methods. The basic element of most neural networks is a combination of linear transformations and nonlinear activations. We believe this linearity itself is redundant and propose a new method for designing a fast neural network by using nonlinear b-spline functions instead of linear multiplications. We propose an end-to-end design flow for high-speed b-spline neural networks, which shows an actual speedup of 2.4 to 4.1 times compared to conventional neural networks when the network is implemented on Raspberry Pi Zero and Raspberry Pi Pico.

Although convolutional neural networks (CNNs) have been widely used in image classification, a large amount of training data is required for training models. Especially in industrial applications, not only is there a demand for reducing the cost of labeling images, but also, it is difficult to obtain a large amount of training images. Several approaches have been proposed to improve classification accuracy by embedding human knowledge other than labels into the models. However, in these approaches, the annotation methods tend to become complex and applicable only to specific model architectures corresponding to those methods.In this study, we propose a new approach to improve the image classification accuracy of CNNs by adding human knowledge as simple region annotations. In our method, annotators draw the regions that attract the most attention as a single ellipse per image for classifying image categories. This annotation is called non-strict attentional region annotation (NARA) and is used for training CNNs together with the labels. We also propose a training method using NARA, which is applicable to common CNNs. For the proof of concept, we have added region annotations to all 5,000 images of the STL-10 training dataset. Through experiments, we demonstrate that our approach is effective to improve image classification accuracy with relatively inexpensive additional cost. Our annotation dataset is available at: https://git.io/stl10nara.

Learning semantically disentangled representations is important for various computer vision tasks, such as image generation and classification. Although it is possible to learn an effective representation in supervised settings, there are problems requiring enormous effort focused in data collection and labeling, and the difficulty in labeling continuously changing events like facial expressions is significant. In this paper, we propose a method for separating the latent representation of facial images into identity factors and facial expression factors using the variational autoencoder (VAE) framework. In our method, we only use subject labels to control training, and we do not use information attached to facial expressions like emotional labels. The separation between extracted facial expression factors and identity features is very useful for controlling image generation and for classifying facial expressions. Using this latent representation, we also suggest a new approach for facial expression recognition with a simple clustering method, which is based on Euclidean distance. Our classification method dramatically reduces the cost of labeling. The experimental results show that our method successfully disentangles the representation of facial images and separates the latent representation into identity and facial expression factors. Moreover, in a facial expression recognition task, our approach shows advantages over the baseline method without supervision.

Although convolutional neural networks (CNNs) have significantly evolved and demonstrated outstanding performance, their uninterpretable nature is still considered to be a major problem. In this study, we take a closer look at CNN interpretability and propose a new method called Evolutionary Generative Contribution Mappings (EGCM). In EGCM, CNN models incorporate both a classification mechanism and an interpreting mechanism in an end-to-end training process. Specifically, the network generates the class contribution maps, which indicate the discriminative regions for the model to identify a specific class. Additionally, these maps can be directly used for classification tasks; all that is needed is a global average pooling and a softmax function. The network is represented by a directed acyclic graph and optimized using a genetic algorithm. Architecture search enables EGCM to deliver reasonable classification performance while maintaining high interpretability. We apply the EGCM framework on several datasets and empirically demonstrate that the EGCM not only achieves excellent classification performance but also maintains high interpretability.

In recent years, there has been substantial research on uncertainty and explainability to improve the reliability of deep learning predictions. In the field of image processing, previous methods can be intuitively understood by humans and can be used to evaluate the reliability of predictions. However, they cannot be applied to complex time series prediction because of the difficulty for humans in evaluating the reliability of predictions. Previous methods focused on either uncertainty or explainability; therefore, the evaluation of the methods and predictions depended only on human knowledge. In this paper, we propose a time series prediction method focusing on both uncertainty and explainability. Our method allows the evaluation of predictions from multiple perspectives by using uncertainty and explainability without relying solely on human knowledge. The explanation obtained from the unimodal distribution prediction model is unreliable because the predictive distribution of time series prediction can be an inherently multi-peaked. The unimodal model does not correspond to the actual phenomenon and does not match with human knowledge. Therefore, we solve the time series prediction as a multi-class classification problem. From the verification of the proposed model with real time series, we confirm the effectiveness of the (1) prediction based on the expected value of the class probability distribution, (2) confidence of the prediction, and (3) series importance based on the confidence. Moreover, we show that our model can improve the robustness of the time series prediction and evaluate its reliability.

Monitoring and predicting fishing activity is important for the fishery resource management and the maritime traffic safety. In this paper, we trained deep learning models by grid images from satellite observations to predict areas of fishing activity in next three-day period. The best model predicted the estimated size of fishing areas with more than 70% coverage for days 1 and 3 after prediction, and areas of congestion with more than 50% certainty for day 1. In particular, the model using time information performed with higher coverage and certainty. The models might be used not only to predict fishing activities for resource management and navigation safety but also to supplement data in cloudy weather for supporting optical satellite observations.

The convolutional neural network (CNN), one of the deep learning models, has demonstrated outstanding performance in a variety of computer vision tasks. However, as the network architectures become deeper and more complex, designing CNN architectures requires more expert knowledge and trial and error. In this article, we attempt to automatically construct high-performing CNN architectures for a given task. Our method uses Cartesian genetic programming (CGP) to encode the CNN architectures, adopting highly functional modules such as a convolutional block and tensor concatenation, as the node functions in CGP. The CNN structure and connectivity, represented by the CGP, are optimized to maximize accuracy using the evolutionary algorithm. We also introduce simple techniques to accelerate the architecture search: rich initialization and early network training termination. We evaluated our method on the CIFAR-10 and CIFAR-100 datasets, achieving competitive performance with state-of-the-art models. Remarkably, our method can find competitive architectures with a reasonable computational cost compared to other automatic design methods that require considerably more computational time and machine resources.

In multi-agent delivery tasks, Dijkstra's Algorithm, which identifies and plans the shortest path based on costs, is very popular. Specifically, using dynamic path costs, namely updating each path cost to reflect the number of agents having visited it over time, is an efficient method for accomplishing multi-agent delivery tasks. Indeed, it can consider the following logic: “a path that is jammed is not likely to be selected in the routing process”. Suppose tasks that can train with similar delivery tendencies in advance, optimizing the update path cost rule set would lead to more efficient agent path planning. Indeed, it would consider that “the paths that will be jammed are not likely to be selected in the routing process”. Thus, we propose a dynamic path cost update method that reflects delivery tendencies. We design the update path cost if-then rule (what path costs should be added, and under what conditions?) as a current partial agent distribution around an observation point for the if-part, and the amount of added (or subtracted) path cost for the then-part. In one delivery task, each observation point is tied to partial agents' distribution. If a distribution is collated with an update rule's if-part, the corresponding path of the observation point is updated given the update path costs that the collated rule's then-part indicates. This update continues during a task at each observation point at an interval. The set of update path cost if-then rules is optimized with Learning Classifier Systems. We evaluated our method with a multi-agent delivery task that simulates a warehouse swarm robot's Pickup and Delivery task. The results showed our method accomplished the tasks faster than a previous approach. In addition, we evaluated the superiority range of the proposed method when the number of agents used changes. Finally, we compared the path cost transition tendency. It revealed our method was more likely to drastically alter path costs.

Over the last few years, tumor detection with deep learning has attracted ample attention, and various models have been proposed. Some models obtain output images directly from end-to-end models. Other models consist of a complicated network structure. However, it is difficult for doctors to understand the detection procedures of these models. This difficulty hinders their practical implementation. In this paper, therefore, we propose a model that detects tumors in a way that is easy for doctors to understand. The proposed deep neural network model was inspired by a way that doctors interpret images. It consists of two processes: tumor position estimation, and tumor/non-tumor classification. To evaluate the proposed method, we used magnetic resonance images actually collected by a hospital. The output images reveal that the proposed method's procedure of tumor detection is consistent with the way that doctors detect tumors. Likely tumor positions are first estimated. Then, these likely positions are narrowed down to the correct tumor positions. Our proposed method detected tumors with 94.14% accuracy and 93.44% recall.

We propose a method for designing convolutional neural network (CNN) architectures based on Cartesian genetic programming (CGP). In the proposed method, the architectures of CNNs are represented by directed acyclic graphs, in which each node represents highly-functional modules such as convolutional blocks and tensor operations, and each edge represents the connectivity of layers. The architecture is optimized to maximize the classification accuracy for a validation dataset by an evolutionary algorithm. We show that the proposed method can find competitive CNN architectures compared with state-of-the-art methods on the image classification task using CIFAR-10 and CIFAR-100 datasets.

Considering the ease of data collection, it is desirable to build a voice conversion system (VCS) from non-parallel voice data, with different sentences read by the source and target speakers. Previous non-parallel VCSs have used either a mel-cepstrum or spectral envelope as an input feature. However, these features have different acoustic characteristics that play important roles in speaker recognition. Thus, we propose a non-parallel VCS that efficiently uses both mel-cepstrum and spectral envelopes as input features. Our method is based on three key strategies: 1) we use generative adversarial networks for voice conversion; 2) we add noise to facilitate the training of the formant part; and 3) we integrate the acoustic features to generate high-quality converted voices. Subjective evaluations in the Voice Conversion Challenge 2016 (VCC 2016) revealed that our model outperformed the previous approaches in terms of the naturalness and similarity of the converted voice.

Multimodal learning has received considerable attention in recent years. Most prevailing approaches use multiple networks to learn a shared representation across modalities. However, these methods are limited to the laboratory scale because of unavailability of multimodal data. Percolative learning is a straightforward framework for solving this problem; we train our model using data from multimodalities. The model performs well during the testing phase, although only data from a single modality is provided. In this paper, we extend this framework to the problem of time-series prediction, specifically with focus on applications on maritime indices. Then, we attempt to exploit the potential of percolative learning. We also attempt to employ a genetic algorithm to find suitable architectures, which is one of the most important ingredients of percolative learning. The experimental results demonstrate the potential of our method and its superiority to other baselines.

The convolutional neural network (CNN), which is one of the deep learning models, has seen much success in a variety of computer vision tasks. However, designing CNN architectures still requires expert knowledge and a lot of trial and error. In this paper, we attempt to automatically construct CNN architectures for an image classification task based on Cartesian genetic programming (CGP). In our method, we adopt highly functional modules, such as convolutional blocks and tensor concatenation, as the node functions in CGP. The CNN structure and connectivity represented by the CGP encoding method are optimized to maximize the validation accuracy. To evaluate the proposed method, we constructed a CNN architecture for the image classification task with the CIFAR-10 dataset. The experimental result shows that the proposed method can be used to automatically find the competitive CNN architecture compared with state-of-the-art models.

In recent years, significant research has been conducted on grasp planning for multifingered robot hands. These studies have focused on determining how to obtain suitable grasps from among an infinite number of candidate grasps. This domain's goal is a successful application to unknown environments through the adoption of the extracted grasps. Under difficult conditions, such as grasping a target object that is adjacent to other objects, manipulating robot hands by indicating grasping points has been insufficient. Instead, grasp strategies that construct movements using each finger's joint servo controls and robot hand movements should be used. In addition, it is necessary to automatically acquire various grasp strategies to apply to unknown environments. In this paper, we propose a method that automatically obtains grasp strategies using a real-coded genetic algorithm (RCGA), which is an evolutionary algorithm. This method derives grasp strategies by optimizing combinations and structures that consist of simple finger joint servo controls and robot hand movements. By applying our method to several objects on a simulator, we collected various grasp strategies capable of handling difficult conditions.

In some of the EEG-based recognition tasks, for example, EEG-based emotion recognition (EEG-ER), enhancing feature extractors is difficult. In such cases, the use of deep neural networks which are capable of classification and recognition by the input of raw data is desirable. Therefore, effective components and models of neural networks for EEG-based recognition must be proposed. In addition, the capability of easily interpreting the feature networks learned is also needed not only from the viewpoint of enhancing features but also from the neuroscientific viewpoint. This paper proposes a discrete Fourier transform (DFT) layer, inverse DFT layer, and other components to compose a frequency filter module. This module was proposed for embedding a bandpass filter suitable for EEG data and the targets, and interpreting features in frequency domain. The proposed models were compared with their counterparts and state-of-the-art models, and evaluated according to their accuracies and visualizing features for person recognition and emotion recognition. The results showed that the frequency filter module is effective in preprocessing or interpreting some features in EEG with a characteristic in frequency domain.

Image segmentation and image recognition are challenging processes, and the methods of merging those two processes like semantic segmentation have been studied. However, it is a lot of labor to construct the processes of segmentation and recognition manually, so automatic construction of those approaches using machine learning or evolutionary computation have been proposed. In this paper, we propose a model of pixel-wise image segmentation and recognition using Cellular Evolutionary Networks (CEN). Our proposed model is composed of a regular array of the identical feed forward networks, represented in Cartesian Genetic Programming (CGP), and each CGP connects with neighbor CGPs. Besides, we also propose a new model of CEN called Three-Dimensional Cellular Evolutionary Networks (3D-CEN), which is composed of multiple CENs. We applied CEN and 3D-CEN to road scene images and verified the effectiveness of our method, and experimental results showed that our new model acquired better performance compared with other methods if efficient evolution for CEN is done.

In this paper, we design a hierarchical feature construction method for image classification. Our method has two feature construction stages: (1) feature construction by a combination of primitive image processing filters, and (2) feature construction by evolved filters. We verify the image classification performance of the proposed method on the MIT urban and nature scene dataset. The experimental results show that the two-stage feature construction improves the classification accuracy compared to single stage feature construction. In addition, the proposed method outperforms several existing feature construction methods.

This paper presents Cellular Fuzzy Oriented Classifier Evolution (CFORCE), a generic method for constructing fuzzy rules to divide an image into two segments: object and background. In CFORCE, a pair of fuzzy classification rule sets for object and background is defined as a processing unit, and the identical units are allocated in each pixel of an input image. Each unit computes matching degree of each pixel with object and background class iteratively with considering the matching degree of neighbor units. The algorithm has mainly two features: 1) designing the fuzzy rules using Fuzzy Oriented Classifier Evolution (FORCE) which develops fuzzy rules represented as directed graphs flexibly and automatically by Genetic Algorithm, and 2) performing iterative segmentation with considering spatial relationship between pixels besides local features. In natural image segmentation, many pixels are overlapped between different clusters. Therefore, considering the spatial relationship is important to classify the overlapped pixels correctly. We applied CFORCE to three different object segmentation, and showed that CFORCE extracted object regions successfully.

Operation video recording is one of efficient methods for analyzing surgical workflow and intraoperative incident detection. In “Intelligent Operating Room” at Tokyo Women's Medical University Hospital, the special neurological surgery called awake craniotomy is recorded by video recording system IEMAS (Intraoperative Examination Monitor for Awake Surgery). There are a number of useful video records of surgical procedure such as patient reactions and surgical operations. However, these surgical event tags which are used for surgcial workflow analysis are not contained in IEMAS records. IEMAS is composed multi-view video cameras, so manual tagging for video records is a lot of labor because of the large length of surgical operation videos. In awake craniotomy, electrical stimulation is one of significant surgical operations for detecting eloquent brain areas. In this paper, we propose the automatic detection method for the stimulation points onto patient's brain areas from raw IEMAS video records by using image processing approach. In the previous work, we proposed stimulation timing detection method from surgery sound records of IEMAS. Hence, we propose a detection method of surgical instrument for electrical stimulation in order to tag the stimulated positions on surgical view. However, detected positions on the video frame coordinates depend on camera view changes. Therefore, we map video frame positions to representative video frame positions by using homography transformation in order to enable to analyze stimulated points relation. We applied automatic video tagging method for several raw IEMAS video records and show its performance.

We have been proposing a disparity propagation-based stereo matching algorithm using cellular evolutionary neural network (CEN). Although our previous work demonstrated our algorithm is able to obtain decent accuracy for various scenes with low computational cost, its accuracy is limited by the simple initial disparity calculation and lack of proper propagation. In this paper, therefore, we propose an initial disparity optimization and disparity diffusion/absorption-based approach. The first feature mainly calculates initial disparities by an evolutionary-optimized matching cost function. The second feature then diffuse/absorb them according to the reliability of the initial disparity by utilizing state transition of CEN. Experimental results show that our new algorithm exceeds the common methods and our previous one with low computational cost, indicating the key features boost accuracy, especially for texture less regions, without being computationally expensive.

It is thought that human being recognizes a complicated figure by combining simple figures. This is "figure alphabet hypothesis" and these simple figures are called "figure alphabet". We considered "the mechanism in which a complicated figure is recognized with the combination of the figure chosen from comparatively simple figure groups", and applies it to a pattern classification. The proposed method assumes the figure alphabet to be the dot pattern (Alphabet Dot Pattern, ADP) of an N × N pixels. Because there are many kinds of ADP, ADP group is optimized by Genetic Algorithm (GA). And, the euclidean distance of an input figure and an ADP group is calculated, and classifies the figure. The proposal technique was previously applied to the classification problem of the binary multifont figure, and the validity was shown. In this research, the result applied to the gradation images. As a result, the classification of the face image and the pedestrian image obtained a high correct answer rate.

Fuzzy ORiented Classifier Evolution (FORCE) is a graph-based genetic fuzzy system which we have previously proposed. FORCE constructs fuzzy classification rules automatically by evolving directed graphs composed of fuzzy conditions using Genetic Algorithm. In this work, to improve FORCE about efficiency of rule optimization and expressiveness of Membership Functions (MFs), we introduce two new ideas into FORCE: Edge Mutation (EM) and Parameter tunable MFs (PMFs). EM changes node connections with considering current graph structure to develop rules efficiently unlike the original mutation changing them just randomly. PMFs are MFs characterized by real-coded parameters optimized using uniform and non-uniform mutation. PMFs improve the expressiveness of MFs, which are represented by combination of user-defined parameters in the previous work. We tested the improved FORCE with 21 classification data sets in comparison with our previous model and a common method, and experimental results showed the proposed ideas improved classification accuracy of FORCE.

The data visualization, which reduces data dimension to make us easy to see data directly, is important in data mining. It is important that one category becomes one cluster (we define this as data cohesion) and the clusters are standoff to each other (we define this as cluster separation) in data visualization. In this paper, we propose a self-organizing method using data movement on a spherical surface for data visualization. The proposed method puts all data points on the spherical surface and each data point moves on the spherical surface under the force from all the other data points. The key features of the proposed method are using a spherical surface as output space and employing weighted inter-point distance which emphasizes similarity between these data points. The experimental results show that the proposed method visualizes data with high data cohesion and high cluster separation by dint of above features.

Image processing and recognition technologies are becoming increasingly important. Automatic construction methods for image transformation algorithms proposed to date approximate adequate image transformation from original images to their target images using a combination of several known image processing filters by evolutionary computation techniques. In this paper, we introduce the adaptive image processing filters that process according to the features of an input image. The processing of the adaptive filters is decided based on the local features of an input image. We implement them to feed-forward genetic image network (FFGIN) that is one of the automatic construction methods for image transformations. Then we apply our method to the problems of segmentation of organs and tissues in medical images. Experimental results show that our method constructs the effective segmentation algorithms that extract multiple regions respectively.

When we evaluate the search performance of an evolutionary computation (EC) technique, we usually apply it to typical benchmark functions and evaluate its performance in comparison to other techniques. In experiments on limited benchmark functions, it can be difficult to understand the features of each technique. In this paper, the search spaces that emphasize the performance difference of EC techniques are evolved by Cartesian genetic programming. We focus on a real-coded genetic algorithm, which is a type of genetic algorithm that has a real-valued vector as a chromosome. In particular, we generate search spaces using the performance difference of real-coded crossovers. In the experiments, we evolve the search spaces using the combination of three types of real-coded crossovers. As a result of our experiments, the search spaces that exhibit the largest performance difference of two crossovers are generated for all the combinations.

Using well-established techniques of Genetic Programming (GP), we automatically optimize image feature filters over several inputs and within transformation images, improving the Automatic Construction of Tree-Structural Image Transformation (ACTIT) system. Our objective is to also produce optimal solutions in substantially less computation time than require for generating features of ACTIT. We improved the algorithm feature filters in the process through GP, which are expressed by trees in Automatic Construction of Tree-Structural Image Transformation, to reduce computation time. Through our experimentation, we show that our new approach is accurate and requires less computation time by maintaining the feature images in conjunction with the original images.

In the fields of image processing and recognition, image segmentation is an important basic technique in which an image is partitioned into multiple regions (sets of pixels). In this paper, we propose a method for evolutionary image segmentation based on multiobjective clustering. In this method, two objectives, overall deviation and edge value, are optimized simultaneously using a multiobjective evolutionary algorithm. These objectives are important factors for image segmentation. The proposed method finds various solutions (image segmentation results) by the use of an evolutionary process. We apply the proposed method to several image segmentation problems and confirm that various solutions are obtained. In addition, we use a simple heuristic method to select one solution from the original Pareto solutions and show that a good image segmentation result is selected.

Currently, various automatic programming techniques have been proposed and applied in various fields. Graph Structured Program Evolution (GRAPE) is a recent automatic programming technique with graph structure. This technique can generate complex programs automatically. In this paper, we introduce the concept of automatically defined functions, called automatically defined nodes (ADN), in GRAPE. The proposed GRAPE program has a main program and several subprograms. We verified the effectiveness of ADN through several program evolution experiments, and report the results of evolution of recursive programs using GRAPE modified with ADN.

Image processing and recognition technologies are required to solve various problems. We have already proposed the system which automatically constructs image processing with Genetic Programming (GP), Automatic Construction of Tree-structural Image Transformation (ACTIT). However, it is necessary that training image sets are properly classified in advance if they have various characteristics. In this paper, we propose Modified Parallel-ACTIT which automatically classifies training image sets into several subpopulations. And it optimizes tree-structural image transformation for each training image sets in each subpopulations. We show experimentally that Modified Parallel-ACTIT is more effective in comparison with ordinary ACTIT.

In recent years a lot of Automatic Programming techniques have developed. A typical example of Automatic Programming is Genetic Programming (GP), and various extensions and representations for GP have been proposed so far. However, it seems that more improvements are necessary to obtain complex programs automatically. In this paper we proposed a new method called Graph Structured Program Evolution (GRAPE). The representation of GRAPE is graph structure, therefore it can represent complex programs (e.g. branches and loops) using its graph structure. Each program is constructed as an arbitrary directed graph of nodes and data set. The GRAPE program handles multiple data types using the data set for each type, and the genotype of GRAPE is the form of a linear string of integers. We apply GRAPE to four test problems, factorial, Fibonacci sequence, exponentiation and reversing a list, and demonstrate that the optimum solution in each problem is obtained by the GRAPE system.

In this paper, the authors propose a fast evolutionary image processing system. The authors employ graphics processing unit (GPU) to automatic construction of tree-structural image transformation (ACTIT) for the purpose of reducing optimization time. Besides, the system calculates in parallel by using multiple GPUs for the fast processing. The optimization speed of the proposed system is several hundred times faster than that of the ordinary ACTIT. Experimental results show that the proposed system is effective.

Recently, we often deal with multi-dimensional data in the field of image processing. Thanks to the development of imaging equipments and computational processing speed, we can process 3D image data. For instance, 3D volume image and video image data have been used in various fields. In general, it is difficult to construct 3D image processing compared with 2D ones. Thus, we previously proposed the method named “3D-ACTIT; Three Dimensional Automatic Construction of Tree-structural Image Transformation” to construct various 3D image processing procedures automatically. This system constructs tree-structural image processing filter from combina tion of one-input one-output filters and two-inputs one-output filters by using Genetic Programming (GP). We give only some training image sets, this system constructs complex 3D image processing automatically. By using this system, we succeeded to construct the complex 3D medical image processing. For instance, the extraction processing of internal organs from 3D PET data and abnormal signals from 3D Diffusion Weighted Image are constructed. In addition to the field of medical image processing, in Intelligent Transport system (ITS) and security system, video image processing requires complex and 3D processing. In this paper, we apply 3D-ACTIT to video image data. We took the video image data using some compact robots on the assumption of in-vehicle camera. 3D-ACTIT constructs the processing which extracts the moving objects.

In this paper, we apply graph structured program evolution (GRAPE) to evolution of general sorting algorithm. GRAPE is a new Automatic Programming technique. The representation of GRAPE is graph structure, therefore it can express complex programs (e.g. branches and loops) using its graph structure. Each program is constructed as an arbitrary directed graph of nodes and data set. GRAPE handles multiple data types using data set for each type, and the genotype of GRAPE is the form of a linear string of integers. The aim of this work is to evolve a program which correctly sort any sequence of numbers. We demonstrate that GRAPE constructs general sorting algorithm automatically.

We consider realistic settings of an artificial market from the viewpoint of a long memory process. With the aim of analyzing the mechanism of stock price change, we construct an artificial stock market composed of multiple agents whose investment strategies are represented by tree-shaped programs. The market is optimized using genetic programming so that the change of its stock price resembles that of a "real" stock market statistically. In order to perform an efficient optimization and analyze agents' behavior easily, we use ADG - automatically defined groups proposed previously. We show experimentally that complex changes in a real market appear in the proposed artificial market.

A new template matching method that is based on marker-controlled watershed segmentation (TMCWS) is presented. It is applied to recognize numbers on special metal plates on production lines where traditional image recognition methods do not work well. In contrast to previous matching algorithms, TMCWS firstly creates a marker image for each pattern, and then takes both the pattern image and its corresponding marker image as a template window and shifts this window across a gradient space or an unknown image pixel by pixel to do a search. At each position, the marker image is used to try to extract the contour of the target object with the help of marker-controlled watershed segmentation, and the pattern image is employed to evaluate the extracted shape in each trial. All the pattern images and their corresponding marker images are tried and the pattern that best matches the target object is the recognition result. TMCWS contains shape extraction procedures. Experiments are performed with this method on nearly 400 images of metal plates and the test results show its effectiveness in recognizing numbers in noisy images.

In real market, the squares of stock price change rates have high autocorrelation, and the change rates show high peak and fat tail distribution. With the aim of analyzing the mechanism of the stock price change, we construct an artificial stock market composed of multiple agents whose investment strategies are represented by tree-shaped programs. The market is optimized by using a Genetic Programming so that the change of its stock price resembles that of "real" stock market statistically. In order to perform an efficient optimization and to analyze agents' behavior easily, we use ADG; Automatically Defined Groups previously proposed by authors. We show experimentally that complex changes such as real market appear in the proposed artificial market. Moreover we analyze the interaction of agents which causes realistic stock price changes.

We previously proposed an automatic construction method of image transformations. In this method, we approximated an unknown image transformation by a series of several known image filters, and a genetic algorithm optimizes their combination to meet the processing purpose presented by sets of original and target images. In this paper, we propose an extended method named "Automatic Construction of Tree-structural Image Transformations (ACTIT)". In this new method, a tree whose interior nodes are image filters and leaf ones are input images approximates the transformation. The structures of the trees are optimized using genetic programming. ACTIT finds practical filter combinations that are too complicated to be designed by hand. It can be applied to various kinds of image processing tasks. We show examples of its applications to document and medical image processing.

A method to approximate the image transformation from an original image to its ideally processed target image by a sequence of several image transformation filters is proposed in this paper. The ideally processed image is assumed to be generated manually by the use of drawing software. In this method, the order of applying image transformation filters is determined by a genetic algorithm. This method can be applied to construction of an expert system for image processing.

A method is proposed to extract two-dimensional shapes which are similar to a given model shape from a binary image composed of black and white pixels. This extraction problem is equivalent to the problem of determining the position, size, and rotational angle of each similar shape in the binary image. The model shape is transformed with four space transformation parameters, (chi) c, yc, M, and (theta) , and the transformed model shape is overlapped with the binary image. Parameters (chi) c and yc denote xy coordinate values of the center of gravity of the transformed model shape, M is magnification ratio and (theta) is rotational angle. The research goal here is to find out the space transformation parameter set that gives the maximum matching rate between the transformed model shape and a similar shape in the binary image. Genetic algorithm (GA), which is a kind of searching or optimizing algorithm, is employed for this problem. In this method, several virtual living things whose chromosomes represent space transformation parameters are randomly generated in a computer, and they are evolved according to GA. As the result of generation iterations, an evolved individual corresponding to the best space transformation parameter set is obtained. Algorithm of the method and several experimental results are described.

A new algorithm is proposed by which the circular arcs or straight lines in the original drawings are transformed into straight lines in a U-V complex plane. After these parameters of straight lines in the U-V plane have been extracted, circular arcs and straight lines in the original drawings are recognized. The results show that the extraction time and memory storage are reduced.>

We propose a new method to extract arbitrary shapes such as lines, circles, ellipses and other complex shapes, from noisy binary images. In this method, shapes in a given binary image are traced by the tracer named PVFT (Pseudo View Field Tracer ). The movement of PVFT is similar to that of the view field of a man who recognizes arbitrary shapes with a restricted view field. That is, PVFT selects line segments of a shape in a noisy image, and traces them. Moreover, movement of PVFT is controlled according to the shape to be extracted.