자기 지도 학습 모델을 사용한 중복 이미지 탐지
Near-Duplicate Image Detection Using a Self-Supervised Learning Model
0
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
중복 이미지 탐지(near-duplicate image detection)는 중복에 대하여 미리 정의된 특정 조건을 만족하는 이미지 쌍을 찾는 문제이다. 최근 연구에서는 중복 이미지 쌍을 찾기 위해 이미지 분석을 위한 딥러닝 기술이 활용되고 있는데, 기존의 딥러닝 기법은 주로 분류 문제를 중심으로 한 지도 학습 방식을 사용하므로 클래스 라벨과 같은 교사 신호가 필요하다. 또한 중복 이미지 탐지와 같이 주어지는 입력 이미지의 도메인이 객체·풍경·인물 등과 같은 특정 그룹에 국한되지 않는 일반적인 경우에는 충분한 일반화 성능을 확보하기 힘들다. 이 문제를 해결하기 위해 본 논문은 최근 주목받는 자기 지도 학습 모델(self-supervised model)을 이용하여 이미지의 특징을 추출하고, 다른 이미지의 특징과 거리를 비교하는 방법으로 중복 이미지 탐지 하는 방법을 제안한다. 본 논문은 California-ND 데이터와 MIR-Flickr Near Duplicate(MFND) 데이터를 사용하여 다른 모델과 비교 실험을 수행하였고, 제안하는 방법이 기존 방법보다 우수한 성능을 내는 것을 확인하였다.
Abstract
Near-duplicate image detection is a find image pairs that satisfy some predefined conditions on image duplication. To solve this problem, deep learning techniques for image analysis are being used actively. The conventional approaches mainly use supervised learning methods for classification tasks, which needs appropriate supervised signals such as class labels. Also, it is difficult to have sufficient generalization performance when the domain of input images is not limited to a specific group. Considering these difficulties, this paper proposes a method for near duplicate image detection by extracting image features using a self-supervised model, which has recently been attracting attention, and comparing distances between the features. We conduct comparative experiments with other models using California-ND data and MIR-Flickr Near Duplicate (MFND) data, and confirm that our method achieves the state-of-the-art performance.
참고문헌 (17)
- Content-based image retrieval and feature extraction: a comprehensive review, "Latif, Afshan", Mathematical Problems in Engineering 2019, ~ (2019)
- Connor, "Identification of MIR-Flickr Near-Duplicate Images", VISAPP 2015, 10th International Conference on Computer Vision Theory and Applications (2015)
- Jinda-Apiraksa, Amornched, "California-ND: An annotated dataset for near-duplicate detection in personal photo collections.", 2013 Fifth International Workshop on Quality of Multimedia Experience (QoMEX) (2013), IEEE
- Connor, Richard, "Quantifying the specificity of near-duplicate image classification functionss", 11th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (2016)
- Morra, Lia, "Benchmarking unsupervised near-duplicate image detection", Expert Systems with Applications, 135, pp.313~326 (2019)
- Heesung Yang, "Deduplication of Retrieved Image Data Using Deep Network Features", 2021 한국정보과학회 학술발표논문집, pp.1518~1520 (2021)
- Gordo, Albert, "European conference on computer vision" (2016), Springer
- Zhang, Yi, "Learning near duplicate image pairs using convolutional neural networks", International Journal of Performability Engineering, 14(1), pp.168~ (2018)
- Chen, Ting, "A simple framework for contrastive learning of visual representations", International conference on machine learning. PMLR (2020)
- He, Kaiming, "Momentum contrast for unsupervised visual representation learning", Proceedings of the IEEE, 2020, ~
- Oliva, Aude, "Modeling the shape of the scene: A holistic representation of the spatial envelope", International journal of computer vision, 42(3), pp.3~ (2001)[DOI]
- Zhou, Zhili, "Near-duplicate image detection system using coarse-to-fine matching scheme based on global and local cnn features", Mathematics, 8(4), pp.644~ (2020)
- Koch, Gregory, "Siamese neural networks for one-shot image recognition", ICML deeplearning workshop, 2 (2015)
- Zhang, Yi, "Single-and cross-modality near duplicate image pairs detection via spatial transformer comparing CNN", Sensors, 21(1), pp.255~ (2021)
- He, Kaiming, "Deep residual learning for image recognition", Proceedings of the IEEE conference on computer vision and pattern recognition (2016)
- Coates, Adam, "An analysis of single-layer networks in unsupervised feature learning", Proceedings of the fourteenth international conference on artificial intelligence and statistics. JMLR Workshop and Conference Proceedings (2011)
- Huiskes, Mark J., "The mir flickr retrieval evaluation", Proceedings of the 1st ACM international conference on Multimedia information retrieval (2008)