KiWi：公的機関における紙書類の安全な電子化を実現する 個人情報管理・請求システムの提案

瀬尾 敦生, 榎本 浩義, 石橋 治樹, 奥河 董馬, 濱田 捷聖

情報処理学会第88回全国大会講演論文集

リアルタイム姿勢推定による弓道フォーム評価手法の提案とWeb実装

奥河 董馬, 益崎 智成, 牧山 隆洋

令和7年度 電気・電子・情報関係学会四国支部連合大会 講演論文集

複素指数カーネルに基づく高速・構造保持型 Retinex手法 XCRの提案

奥河 董馬

Jxiv

本研究では，低照度画像において構造保持と高速性を両立した新しいRetinexベースの強調手法「eXponential-Cos Retinex（XCR）」を提案する．従来のSingle-Scale Retinex (SSR) や Multi-Scale Retinex (MSR)，およびLIMEなどの手法では，ノイズ増幅やハロー発生，色ずれ，計算コストの高さといった課題が指摘されてきた．XCRは，複素指数関数と余弦関数に基づく新しい分離型カーネルにより，構造の破壊を抑えつつ照明成分を安定的に推定する．また，照明値に応じた適応的な反射成分増幅制御とγ補正により，過度な明部変化を抑えながら自然な画像表現を実現する．提案手法は，従来手法よりも高い構造類似度指数（SSIM）と高速な処理時間を達成し，特にリアルタイム性能に優れている．実験結果より，XCRはRetinex系手法の新たな選択肢として有望であることが示された．

Fast Single-Scale Retinexの提案と低照度環境におけるリアルタイム骨格検出性能の評価

奥河 董馬

Jxiv

本研究では，従来困難であった低照度環境下における骨格検出を実現するため，Single-Scale Retinexを拡張したFast Single-Scale Retinexを提案する．本手法では，画像から輝度成分のみを抽出し，対数空間で照明成分を分離することで明るさの非均一性を補正する．骨格検出にはMediaPipeを使用し，955フレームの映像を対象に評価を行った．その結果，補正後の平均二乗誤差（MSE）は 472.73 となり，暗部の階調が有効に拡張されたことで，従来検出が困難であった指先や関節位置の検出が可能となった． 本手法は，低照度下での人物解析や行動認識への応用に有効であることを示す．

KiWi: A Secure Information Request and Management System for Public Institutions through the Digitization of Paper-Based Personal Data

Atsuki Seo, Hiroyoshi Enomoto, Haruki Ishibashi, Toma Okugawa, Shousei Hamada

Proc. 88th National Convention of IPSJ

Proposal and Web Implementation of a Kyudo Form Evaluation Method Based on Real-Time Pose Estimation

Toma Okugawa, Tomonari Masuzaki, Takahiro Makiyama

2025 SHIKOKU-SECTION JOINT CONVENTION RECORD OF THE INSTITUTES OF ELECTRICAL AND RELATED ENGINEERS

A High-Speed and Structure-Preserving Retinex Method Based on a Complex Exponential Kernel: XCR

Toma Okugawa

Jxiv

This paper presents eXponential-Cos Retinex (XCR), a new Retinex-based enhancement method designed to achieve both structure preservation and high-speed processing for low-light images. Conventional approaches such as Single-Scale Retinex (SSR), Multi-Scale Retinex (MSR), and illumination-map estimation methods including LIME often suffer from noise amplification, halo artifacts, color distortions, and high computational cost. The proposed XCR introduces a novel separable kernel analytically derived from complex exponential and cosine functions, enabling stable illumination estimation while effectively preserving structural information. In addition, adaptive reflectance amplification based on illumination values combined with γ-correction suppresses excessive changes in bright regions and produces more natural visual results. Experimental evaluations demonstrate that XCR achieves higher Structural Similarity Index (SSIM) and significantly faster processing times than existing Retinex-based methods, exhibiting strong real-time performance. These results indicate that XCR is a promising new alternative within the Retinex family for low-light image enhancement.

Proposal of Fast Single-Scale Retinex and Evaluation of Real-Time Human Pose Estimation Performance under Low-Light Conditions

Toma Okugawa

Jxiv

In this study, we propose the Fast Single-Scale Retinex, an extension of the conventional Single-Scale Retinex, to achieve human pose estimation under low-light conditions, which has been challenging with existing methods. The proposed method extracts only the luminance component from an image and separates the illumination component in the logarithmic domain to correct non-uniform brightness. For pose estimation, we used MediaPipe and evaluated our method on a video comprising 955 frames. As a result, the mean squared error (MSE) after correction was 472.73, and the tonal range in dark regions was effectively enhanced, enabling accurate detection of fingertips and joint positions that were previously difficult to detect. This method demonstrates effectiveness for applications in human analysis and behavior recognition under low-light conditions.