| 藤田 貴大 | 1351093: M, 2回目発表 | 杉本 謙二, | 池田 和司, | 松原 崇充 | ||
|
title: Security Enhancement of Network Control Systems with Secured Control
abstract: We propose a method to enhance security of networked control systems using cryptography. Owing to computation outsourcing tecniques, the proposed method can conceal not only signals in a control system, but also parameters in a controller such as control gains. language of the presentation: Japanese | ||||||
| 真木 勇人 | 1351096: M, 2回目発表 | 中村 哲, | 池田 和司, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title:Probabilistic Enhancement of EEG Component Using Prior
Information of Component-Related Spatial Correlation
abstract: The signal-noise ratio of EEG is very low, which presents serious problems for interpretation and analysis of signals from EEG recordings. Synchronous addition of trials to cancel out background noise or rejecting trials contaminated by eye blinks or other artifacts cause substantial data loss. Therefore, a technique to separate each component of EEG observations is in high demand. Independent component analysis (ICA) is widely used for the purpose. It works well when the number of sensors is equal to or larger than the number of sources. However, the assumption is questionable in the context of EEG signal separation because we do not know the effective number of statically independent brain signals contributing to the EEG. In the field of audio source separation, a technique to enhance objective components using a probabilistic model and multi-channel Wiener filters has been proposed. The method assumes that the amplitude of each component follows a complex Gaussian in each slot of the time-frequency domain, and thus the amplitude of an observed signal follows a Gaussian mixture model (GMM). Parameters including spatial correlation matrices dependent on each component. These are estimated using the EM algorithm to maximize the likelihood of an observation signal. Applying this scheme to EEG signal component enhancement, we set a prior distribution to spatial correlation matrices. Compared to previous work, this allows us to reduce the degree of freedom in parameter estimation, improving estimation performance. Finally, an experiment was carried out, which demonstrated the effectiveness of the proposed approach. language of the presentation: Japanese | ||||||
| 蓑田 由花里 | 1351105: M, 2回目発表 | 中村 哲, | 池田 和司, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title:Measurement of brain activity for cognitive training using NIRS
abstract: Social communication skills is important in the daily life. People with Autism Spectrum Disorder (ASD) have difficulties in these skills. Social skills training is a useful method to improve social communication. However, there is no method to evaluate its objective improvement. I propose a evaluation method for measuring brain activity in pre- and post-training using NIRS. In this presentation, I present the relatinship between autistic traits and brain activities as a preliminary work. language of the presentation:Japanese | ||||||
| 小田垣 佑 | 1361002: D, 中間発表 | 中村 哲, | 池田 和司, | 戸田 智基, | Sakriani Sakti, | Graham Neubig |
|
title: An influence that sense of incongruity gives to event-related potential
Now that cognitive process of Language gradually revealed by comprehensive study, we have a chance to detect mismatch feelings by measuring brain activity. The effect of mismatched words on the sentence comprehension process has been studied for Western languages (Kutas et al., 1980). In a previous study, if a sentence has a word that violates semantic or world knowledge violation word, N400 ERP component appears soon after the reading the violation word (Hagort et al., 2004). In these studies, ERP analysis were based on a category of stimulation prepared in advance. In this study, we presumed that the ERP response based on the subjectivity of the participant was different from that of subjectivity and the word of prepared category. In these experiments, we presented three kinds of sentences as visual stimuli. The first ones were completely correct sentences, the next ones included a word that violated world knowledge, and the last ones included a word that violated semantics. Each sentence was split into segments. In the experiment, each segment was shown to participants for 0.5 seconds with 0.5 second intervals while in spoken stimuli experiment and 1 second intervals between the sentences. After reading each sentence, participants answered a questionnaire which evaluated sentence whether the sentence was correct or not. We calculated ERP results in ordinary way and based on result of questionnaires and compared with them. language of the presentation: Japanese | ||||||
| 片桐 敬太 | 1351030: M, 2回目発表 | 横矢 直和, | 向川 康博, | 佐藤 智和, | 中島 悠太, | 河合 紀彦 |
|
title: Texture Selection for Visually Natural Arbitrary Viewpoint Image Generation
abstract: Arbitrary viewpoint image generation is a technique to synthesize a novel view image from multiple input images capturing an object. Recent research effort have been dedicated to methods that synthesize a novel view by view-dependent texture mapping (VDTM). It applies the suitable input images as a texture to 3D geometry of the object reconstructed automatically by using a multi-view stereo (MVS) technique. However, such a technique often fails in reconstructing 3D geometry accurately, resulting in discontinuity in texture boundaries. This work proposes a method that reduces the discontinuity by optimizing the boundary. In this presentation, I describe the proposed method in detail, as well as the work progress. language of the presentation: Japanese | ||||||
| 篠本 渉 | 1351050: M, 2回目発表 | 横矢 直和, | 向川 康博, | 佐藤 智和, | 河合 紀彦, | 中島 悠太 |
|
title: Distortion Compensation for Novel View Synthesis Using Input Image Features
abstract: Novel view synthesis has recently attracted attention because it can show a user an image from an arbitrary viewpoint different from original camera positions and offer him or her higher immersion feeling. Conventional approaches perform 3D reconstruction in advance and render a scene with view dependent geometry. However, they often produce distortions in some parts where the accuracy of the reconstructed geometry is low. We aim to suppress image distortions by depth map optimization using texture features in input images as constraints to generate a novel view with high quality. In this presentation, we report experimental results in the case of introducing a linearity constraint as an image feature and future issues. language of the presentation: Japanese 発表題目: 原画像の特徴を利用した自由視点画像の歪み抑制 発表概要: 自由視点画像生成は，テレプレゼンスなどの応用において，カメラで撮影された視点以外の任意視点からの画像をユーザに提示し，高い臨場感を与えることができるため近年注目されている． 従来から対象シーンをあらかじめ三次元復元した上で，視点位置に応じたジオメトリを利用する自由視点画像生成が行われているが，三次元復元の精度が低い場合には生成画像に歪みを生じさせる問題がある． そこで，本研究では，入力画像中のテクスチャ特徴を用いた制約によるデプスマップの最適化により画像の歪みを抑制し，高品質な自由視点画像生成を目指す． 本発表では，画像の特徴として直線性に関する制約を導入した場合の実験結果および今後の課題を報告する． | ||||||
| 林 佑亮 | 1351089: M, 2回目発表 | 横矢 直和, | 向川 康博, | 佐藤 智和, | 河合 紀彦 | |
|
title: Video Super-resolution based on the 3D Reconstruction of the scene
abstract: With the increase of the resolution of display devices, techniques for generating a high-resolution video from a low-resolution video are required. Among such techniques, example-based super-resolution methods have been investigated, which super-resolve a target low-resolution video is super-resolved using example high-resolution images. It is necessary for these methods to construct a database including suitable reference images because the effect depends on the examples. In this study, we propose a method to generate a high-resolution video using the reference images that are generated by warping input video and web images based on the 3D geometry reconstructed from the video and web images. Specifically, we first estimate a 3D model of the scene and camera poses from input images. Second, the images are warped to the viewpoint of the target frame using the estimated 3D models and camera poses for reference images. Last, the target frame is super-resolved by energy minimization using the reference images that are selected due to their suitability for increasing the resolution. In the experiment, we confirmed that the proposed method successfully reconstructed the high-frequency component if suitable reference images were selected. In this presentation, I describe the proposed method, the experimental results, and our future issues. language of the presentation: Japanese 発表題目: シーンの三次元復元に基づく動画像の高解像度化 発表概要: 近年，映像表示デバイスの高精細化に伴い，映像コンテンツを高解像度化する技術が必要となっている．動画像の高解像度化手法の一つとして，参照画像を用い低解像度動画像を高画質化する「事例参照型超解像」手法が研究されている．この手法では，高解像度化の効果は入力した参照画像に依存するため，適切な参照画像データベースを構築する必要がある．本研究では，同一シーンを動きながら撮影した動画像およびインターネット上等にある静止画像を入力としてシーンを三次元復元し，これを利用し入力データ中の他画像を変形し参照画像として用いることで，動画像を高解像度化する手法を提案する．具体的な処理として，まず，入力画像群からシーンの三次元モデルを復元し，カメラ位置姿勢を推定する．次に，復元に用いた画像を注目フレームの視点画像に変換し，参照画像とする．その後，参照画像中から高解像度化に適した画像を選択し，エネルギー最小化の枠組みを用いて高解像度化を行う．映像生成実験の結果，適切な参照画像が選択された場合において低解像度画像の高周波成分を復元できることを確認した．本発表では，提案手法の概要，映像生成実験の結果，今後の展望について報告する． | ||||||
| 仲田 昌司 | 1351078: M, 2回目発表 | 小笠原 司, | 向川 康博, | 高松 淳, | 池田 篤俊 | |
|
title: Representation for robust and effective color information processing
abstract: A mobile robot is assumed to perform under many type illumination. In robot vision, it is difficult to identify one color in spite of illumination color change. So I research how to represent color information robustly for color changing. I use sparse coding that is similar to human's early vision and try to separate Light-source color and Object color. language of the presentation: Japanese | ||||||