지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학위논문
- 저자정보
- 지도교수
- 이종태
- 발행연도
- 2023
- 저작권
- 동국대학교 논문은 저작권에 의해 보호받습니다.
이용수1
이 논문의 연구 히스토리 (3)
- 이 논문의 후속연구가 궁금하신가요?
- 연관 학술논문 또는 학술발표를 통해 보다 발전된 연구결과를 확인하실 수 있습니다.
- 이 논문의 연구 히스토리 확인하기
초록· 키워드
오류제보하기
The purpose of this paper is to propose a document image enhancement model based on the vision
transformer model to solve the noise problem that negatively affects document image processing. To this end, we proposed a model consisting of three modules: a damaged encoder, a clean encoder, and a decoder, and performed an experiment on a noisy document image binarization benchmark data set to verify the proposed method. As a result of quantitative and qualitative analysis, it was proved that the document image enhancement performance of the proposed method was improved compared to the existing method, and that the latent representation utilization of the image helps to encode only important information. In addition, it was confirmed that reducing the difference between latent representations had a positive effect on the improvement of the final prediction result. The contribution of this study is as follows. First, this study proposed a latent space utilization method to document image enhancement. Second, this study demonstrated that the performance of the model is improved by using the latent representation of the existing document image enhancement model. Third, this study verified the generalization and scalability of the proposed method by demonstrating that performance is improved by simply adding a latent representation loss function to a general model. It is expected that this study can bring positive results to document image analysis and recognition.
transformer model to solve the noise problem that negatively affects document image processing. To this end, we proposed a model consisting of three modules: a damaged encoder, a clean encoder, and a decoder, and performed an experiment on a noisy document image binarization benchmark data set to verify the proposed method. As a result of quantitative and qualitative analysis, it was proved that the document image enhancement performance of the proposed method was improved compared to the existing method, and that the latent representation utilization of the image helps to encode only important information. In addition, it was confirmed that reducing the difference between latent representations had a positive effect on the improvement of the final prediction result. The contribution of this study is as follows. First, this study proposed a latent space utilization method to document image enhancement. Second, this study demonstrated that the performance of the model is improved by using the latent representation of the existing document image enhancement model. Third, this study verified the generalization and scalability of the proposed method by demonstrating that performance is improved by simply adding a latent representation loss function to a general model. It is expected that this study can bring positive results to document image analysis and recognition.
목차
제1장 서 론 1제1절 연구의 배경 및 목적 1제2절 연구의 방법 및 구성 5제2장 연구에 관한 이론적 배경 7제1절 문서 이미지 강화(Document Image Enhancement, DIE) 7제2절 Vision Transformer 모델 9제3절 잠재 공간의 활용 11제3장 제안 방법 13제1절 제안 방법의 아키텍처 13제2절 제안 방법의 세부 프로세스 151. 사전 학습 152. Damaged Encoder 163. Clean Encoder 164. Decoder 17제3절 손실 함수 20제4장 실험 결과 22제1절 실험 설계 221. 데이터 셋 222. 구현 세부사항 233. 평가 척도 25제2절 정량적 분석 결과 29제3절 정성적 분석 결과 33제4절 모델 확장성 실험 36제5장 결론 39참 고 문 헌 41ABSTRACT 50
최근 본 자료
댓글(0)
0