Historic images attained int the past undergo color deformation and distortion trough time and in the process of acquisition and preservation . We have applied a computer vision algorithm which has shown the best re-colorization results and proceeded an evaluation test.

We have digitalized the analog image through scanning and applied LIME & BM3D algorithm for noise reduction and quality enhancement.

Guided Filtering was applied for the image deblurring process.

CNN trained with paired/unpaired datasets was developed to produce super resolution images.

For immersive VR/AR contents, we have went through state-of-the-art technologies that meet the highest ends of attaining and generating the 2D/3D assets. Also standards for the acquring environment and equipments were established.

A robot arm technology was used for a safe and stabilized digital heritage acquisition environment. The robot’s arm elevates itself systemically and can be programmed for specific positioning.

By establishing standard guidelines in the process of acquiring cultural heritages, curators can directly use and review the museum content in the future.

Basic data conversion technology for cultural heritage asset generation and database linking system was developed.

Development of visualization technology including data composed of 3D ultra-polygon forms. 2D asset visualization, 3D asset visualization, 3D center point modification, and rendering enhancement were performed.

The virtual space exhibition of cultural heritage requires information of space arrangement, characteristics of cultural relics and other environmental details. In order to automatically generate virtual exhibitions according to the characteristics of the relics, technology based on artificial intelligence was developed to collect and visualize the cultural assets in the actual exhibition spaces.

Data acquisition process for creating asset standards(177 attained → 129 transformed into assets), and cultural heritage archive photograph digitalization(8,353 photos) were done. Research on establishing acquisition condition standards for future content utilization was progressed.

Tagging feature keywords of the relics from the e-museum of National Museum of Korea was done for extracting relation traits for cultural heritage search

For the automization of classifying and tagging information of the cultural heritage assets, A language model specialized for analyzing cultural coprus were needed. Given different learning rates and random seeds, the optimal language model was applied in this process.

Using the result of the NER based on the cultural heritage assets, relation network analyzation process was done. We have researched on 4 Korean language models, and worked to find out the best model for analyzation.

Based on Style Transfer Technology, we have supplemented the insufficient training image data and worked on object detection with transfer learning for natural images and virtual DB.

We have worked on to create a platform which users can conveniently use the R&D results for the use of cultural heritage. Establishing databases and utilities for efficient management of cultural heritage digital data and R&D modules were done.

Relationship-based searches of cultural heritage and visualizing relation graphs were implemented in the system so that contents for exhibition education materials can be easily accessed.

To implement a curator-friendly search/view platform, we have analyzed and classified the Korean natural language for effective search of cultural heritage and linked the information to the database. Image tagging and filtering by words were implemented in this process.

To provide the function of converting and sharing the cultural heritage data according to the type and purpose of use, we have development of a GUI module for user convenience and DB design for data generation/management using the cultural heritage asset generation guideline.

Various types of relational visualization methods were studied to express the shape of the relationship using the results of the language model-based relationship extraction technology.

The platform allows cloud utilization of large amounts of data, dualization of storage space, and search technology for meaning-based search.

Created contents for virtual museum using the digital cultural heritage assets (7 demo contents)

Web contents for virtual exhibitions based on digital archiving and intelligent analysis was created. Various web contents can be made from the web platform.

To utilize realistic contents of cultural heritage asset, we have developed a interactive culutral heritage experience content including user information tracking technology. By developing a platform capable of supporting multi-devices, we can experience various interactions with multiple users in multi-devices.

인공지능 기반의 문화유산 영상 이미지 고해상화 변환. 4x SR일 때, PSNR: 27.421, SSIM, 0.7978 획득

문화유산 표준속성에 대한 정의를 재질/빛에 대한 반사등으로 제한, 표준 물리 기반 렌더링(PBR) 템플릿 30종 구성

2D 기가 픽셀 구축을 위한 다중 2D 이미지 데이터 획득 가이드라인 및 고해상도 3D 모델링 데이터 생성을 위한 다중 2D 이미지 데이터 획득 가이드라인 제시

기존의 시간적 노력과 촬영자의 숙련도에 따른 품질 차이 개선을 위한 자동장치를 활용한 최초 개발 시도

국립중앙박물관의 큐레이터, 보존과학부, 교육부서 등의 요구 사항

기반 문화유산 애셋 스키마 정의 및 시스템 설계

원본데이터에서 애셋으로 변환하거나, 애셋 간의 변환을 가능하게 하는 모듈 개발로 데이터 호환성 증대

애셋의 가시화를 활용해 큐레이터들의 요구사항인 직관적 데이터 관리 인터페이스 제공

딥러닝 기반 Multi-object Detection 네트워크 분석 및 적용,

COCO dataset 학습 기반 2D 개체 검출 기술 적용

큐레이터의 요구사항을 반영, 이뮤지엄 문화유산 데이터 분석.

텍스트 데이터 속성 18종 선정, 이미지 데이터 속성 26종 선정

개체명 인식기를 이용한 문화유산데이터의 속성 추출

딥러닝 기반의 개체명 인식기술 이용

문화유산의 검색을 위한 말뭉치와 이미지 데이터셋 구축 도구 개발

웹 기반의 어노테이션 도구로 업데이트와 접근성 확보

지식 관계망 구축을 통한 기록물 데이터 새로운 접근 방향 제시 및 시각화

애셋관리자, 큐레이터, 일반사용자 등으로 구분하고 이를 설계에 반영

고해상도 이미지, 고품질 3D 데이터에 대한 웹에서의 디스플레이를 위한 엔진 구성 및 GUI설계

홀로렌즈2를 활용한 MR기반 문화유산 해설 및 체험 콘텐츠 개발