矢田 紀子

The construction of image transformation manually requires large amounts of effort, and thus several methods of automating it with machine learning, such as neural networks or genetic programming, have been proposed. Most of these methods are just constructed image filters that calculate an output value from values in the local area of each pixel independently. However, in several tasks, such as area detection, information on more distant areas is helpful to processing. In this paper, we introduce a new neural network model for automatic construction of image transformations. The proposed model is composed of a regular array of identical evolutionary neural networks, the Real-Valued Flexibly Connected Neural Networks (RFCN) that we previously proposed, and each RFCN is connected to neighboring RFCNs. The proposed model is called Cellular RFCN (CRFCN). Because of the local connections, each RFCN can consider information on distant areas indirectly. We apply CRFCN to three image transformation tasks, compare it with other methods, and examine its effectiveness. © 2013 Wiley Periodicals, Inc.

近年,コンピュータグラフィックス(CG)技術が高度に発達し,様々な質感をよりリアルに再現することが可能となってきた.しかし,CGで作成された画像に含まれたどのような特徴量が質感認識に関わるのかは,判明していない.本研究では,質感の中でも金属感に注目し,質感知覚に関わる特徴量の推定と定量化を試みた.まず,CGソフトを用いて反射,拡散反射,光沢などのパラメータを様々に設定して画像を36枚作成し,5段階の評定尺度法を用いて金属感に対する主観評価実験を行った.次に,評定尺度法を用いた主観評価実験の結果に対してCGパラメータと金属度の関係について検証し,数量化理論I類を用いて予測式を求めた.さらに,一対比較法を用いた主観評価実験を行い,その結果と評定尺度法から算出した定量化式との比較を行った.

本研究の目的は,人間の視覚特性であるカテゴリカル色知覚と色恒常性を考慮した色認識モデルを獲得し,これを用いて画像中の物体色の色認識を行うことである.画像中の色認識はコンピュータビジョンに非常に重要であるが,従来のカテゴリカル色知覚と色恒常性を備えたカテゴリカル色知覚モデルは画像に対する認識精度が低かった.その原因として,従来モデルにおけるカテゴリ毎の学習量の偏りとカテゴリ境界付近の色に対する認識精度の低さがあげられる.そこで本論文では,カテゴリカルカラーネーミング実験の結果から作成する教師データのカテゴリ毎の学習量の偏りの均一化と,カテゴリ境界付近の色の認識にふさわしいデータ構造への変更を行い,新たなモデルを獲得する.そして,提案モデルに対してカテゴリ毎の認識精度の検証と,カテゴリ境界付近の色の認識精度の検証を行い,その色認識精度が従来のモデルより高いことを示す.また,提案モデルを画像に適用し,画像に対する色認識精度が高いことを示す.

To find and treat diseased part of intracerebral, doctors attempt to understand sterically intracerebral activity from cross-sectional view. This paper proposes a projection mapping method for representation of the intracerebral activity, that is achieved by making the 3DCG based on the image of PET and MRI.

This paper proposes an automatic generation method of marker designs using genetic algorithm. The method aims to create a marker which has high-visibility and large amount of information. An accuracy of decoding is also considered in making the valuation.

Display devices capable of presenting stereoscopic images have become very popular, but the color gamut of the device is insufficient because of using the three RGB primary colors. This study propose a three-dimensional image display device wide color gamut.

This paper proposes a method of 3D shape measurement for translucent object. A translucent object is difficult to measure the 3D shape because of its permeability which causes light scattering. Therefore, this research is aimed at measuring 3D shape of translucent object by reconsidering existing method.

There are many transparent or translucent objects in real world, for example glasses and plastics. However most of image recognition techniques use the object color information and these can't apply to transparent objects. This paper proposes a method for detection of transparent objects from image using texture analysis technics.

This paper proposes a method of recognition of object colors under varying illumination condition in videos. The recognition function is the categorical color perception model using a neural network. The object regions are detected using the depth information from RGB-D camera.

When we shoot a particular person from a fixed position, it is difficult to catch the person always in frames due to occlusion and so on. This paper proposes a method to automatically extract scenes that particular person is reflected from videos shot from multiple directions while tracking other person.

This paper represents a degradation of real objects based on 3D models that made from the RGB-D camera. We obtain the depth information from the RGB-D camera, and create 3D models of real objects. We apply degradation map to 3D models, and superimpose 3D models to real objects.

The purpose of this study is to construct metallic texture space for understanding the space easily as a color space. The results of subjective evaluation are analyzed using principal component analysis for deciding the axis of the metallic texture space.

In the study of Augmented Reality, an object is desired to generate a new object image with different colors from an original image but do not alter the gloss of the object. This research proposes a method of color alterati on of real object using Augmented Reality.

In general, it is difficult to take large and precise photographs of historical materials under good conditions. This paper proposes a method of high-resolution image synthesis using images with different photographed conditions.

Nishiki-e printings has a part of overprinted colorants. It is necessary to represent this part accurately for digital reproduction of Nishiki-e printings. This study proposes a color estimation model of overprinted colorants from spectral reflectance of single color.

Actual printing of Nishiki-e using printing blocks is impossible because the printing blocks are valuable historical things.Therefor a disital reproduction method for printing Nishiki-e is necessary.This paper proposes a reproduction method of Nishiki-e using photometric stereo method.

Nishiki-e is one of the existence to represent the color culture of Japan. This paper proposes a method of automatic color correction and quantitative orange color analysis of Nishiki-e printings.

This paper describes a method of texture reproduction of realistic metal surfaces with age deterioration for applying in computer graphics. We propose a new simulation method for the peeling of preservative coatings of surfaces; the method considers the object geometry and surrounding environment. We also propose a method to decrease processing time using probability distribution maps to speed up the simulation.

Many methods that construct image classification algorithms automatically using evolutionary computation have been studied. Although these classifiers are very effective, several problems have been pointed out. For example, it is difficult to analyze or modify classifiers, and they are too complicated for humans to understand. In this paper, we propose a new method for classification using an evolutionary decision network (EDEN) that emphasizes good human-readability. EDEN automatically constructs an adequate network for classification by combining simple nodes using evolutionary computation. This network is composed of a set of nodes that changes the branches of the decision flow in accordance with the feature values of the input data. We build an effective classifier by optimizing a set of nodes and their threshold values for branching. In experiments, we evaluate EDEN by applying it to image classification problems. The experimental results show EDEN is able to build an effective classifier with a human-readable structure and achieve satisfactory performance.

The seamless integration of real and virtual objects is required for mixed-reality applications. To achieve this goal, we should represent an effect of light reflection like shading, shadowing and inter-reflection between the real and virtual objects. To represent these effects, we have to estimate reflectance of the real objects. The reflectance can be estimated with color and geometry of the objects and light condition of the scene. To calculate at an interactive frame rates, the light sources are distributed on a surface of a dome above the scene. To estimate the reflectance more accurately, we have to calculate distance from the objects to the light source. Therefore, this paper proposes a method to estimate the distance from the objects to the light source and the reflectance of the objects at an interactive frame rates. In the proposed method, two cameras and a marker with a spherical mirror are used. We can use an RGB camera and an IR camera of Microsoft Kinect sensor as the cameras. In other words, by use of the proposed method, we can estimate the distance and reflectance by using the Kinect and the marker with the spherical mirror. In the method, intersection points of reflection vectors on the spherical mirror at each camera are evaluated and the point which has the maximum evaluation value is regarded as an estimation value of the light source position. With the proposed method, we can estimate the light source position and reflectance of the real objects at an interactive frame rates by use of the Kinect and the marker with the spherical mirror, so that we are able to apply the method to various mixed-reality applications.

This paper proposes an augmented reality marker that can be robustly detected even on a cylindrical surface. The marker enables the surface normal estimation of a cylindrical object to realize the presentation of appropriate virtual information on the object. Conventional markers have difficulty detecting and obtaining accurate surface normal in the presence of occlusion or distortion of the marker in the image. Furthermore, it is difficult to identify a feature on a cylindrical object on which to position a marker. These problems are resolved by relying on the characteristic that a line parallel to the central axis of the cylinder maintains linearity. In addition, surface normal is calculated by estimating the object's shape by using transformation matrices. © 2013 IEEE.

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. © 2013 IEEE.

This paper proposes a new design of overlayable transmissive color marker. Each marker is transparent and the marker has different colors, and outside of the marker is a region for estimating order of the overlay. ID patterns are detected from overlaid markers by splitting an input image in red, green and blue channel image. The overlaid order is estimated with the use of average color in the estimation area of overlaid order. Moreover this marker can estimate the relative degree of rotation angle between markers. To evaluate the performance of the proposed marker, we measured the misalignment which can be permitted when detecting each marker and the accuracy of estimated relative angle of the rotation. Moreover the maximum angle which can detect each marker between camera and marker is measured.

The purpose of this study is to get a color perception model of dichromat. We construct a color perception model of dichromat, and analyze the mechanism. Then we have prospect of find human color perception mechanism. We expect it when we understand mechanism of the color perception of the human beings by construct and analysis a dichromat's color perception model. We construct a color perception model of color defects based on the results of psychophysical experiments with optimizes the structure of neural network using genetic algorithm (GA). The neural network used in this paper is real valued flexibly connected neural network (RFCN). RFCN, the evolutionary neural network we previously proposed, is a model that can have high performance in various fields. In RFCN, the structure of the network and the parameter are optimized automatically and flexibly with GA according to tasks we give. So we can obtain the network that has desirable performance without special knowledge about the task. We developed a model that can operate similarly to dichromat's categorical color perception. The results showed that the obtained neural network has similar characteristics to those of dichromat's vision system.