Structural Image De-Identification for Privacy-Preserving Deep Learning

Ko, DongHyun; Choi, Seok-Hwan; Shin, Jinmyeong; Liu, Peng; Choi, Yoon–Ho

doi:10.1109/access.2020.3005911

Cited by 11 publications

(7 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The work discussed by [Zhang et al, 2017] have constructed a framework that can offer privacy while performing image service over cloud environment. The work carried out by [Ko et al, 2020] have used deep learning scheme towards perform de-identification of a structural image. The idea of this work is to prove the accuracy for classification process is like that of deep learning performance without using encrypted image.…”

Section: Literature Reviewmentioning

confidence: 99%

“…However, adopting encryption techniques comes at the cost of a computational burden. Currently, various study approaches are being implemented to ensure privacy protection for the images [Boulemtafes et al, 2020;Cunha et al, 2021;Liu et al, 2020; P. . However, existing approaches are characterized by various shortcomings, forcing the research community to keep exploring more robust solutions.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Analytical and Integrated Framework for Privacy Presrvation of Images in Cloud Ecosystem

Dule¹,

R²

2023

INDJCSE

View full text Add to dashboard Cite

Privacy preservation is always a potential concern while storing or transmitting images with sensitive information within them. A review of existing privacy preservation techniques shows the potential usage of deep learning, adoption of highly sophisticated encryption, lack of packetization scheme, and focus on a singular attack. Therefore, the proposed study presents a novel analytical framework for the privacy preservation of image data in the cloud environment. A novel and simplified mathematical model with layer-based operation is constructed, which facilitates appending image chunks with digital security code and encryption token using a hash to generate a hash vector of an image with a higher degree of secrecy. An extensive analysis is carried out over different datasets of images to find that the proposed scheme offers consistency in its performance and a significant balance between security and computational performance in contrast to frequently exercised privacy preservation schemes. The study outcome exhibited 59% overhead reduction, 25% enhanced signal quality, 65% better resolution, and 40% faster execution time in contrast to the existing scheme.

show abstract

Section: Literature Reviewmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Analytical and Integrated Framework for Privacy Presrvation of Images in Cloud Ecosystem

Dule¹,

R²

2023

INDJCSE

View full text Add to dashboard Cite

show abstract

“…Our approach is most closely related to prior focused on achieving privacy through lightweight encoding schemes. [Ko et al, 2020, Tanaka, 2018, Sirichotedumrong et al, 2019 have proposed de-identification techniques to carefully distort images to reduce their recognition rate by humans while preserving the accuracy of image classification models. Unfortunately, such methods do not offer privacy against realistic attacks.…”

Section: Related Workmentioning

confidence: 99%

NeuraCrypt: Hiding Private Health Data via Random Neural Networks for Public Training

Yala¹,

Esfahanizadeh²,

Oliveira³

et al. 2021

Preprint

View full text Add to dashboard Cite

Balancing the needs of data privacy and predictive utility is a central challenge for machine learning in healthcare. In particular, privacy concerns have led to a dearth of public datasets, complicated the construction of multi-hospital cohorts and limited the utilization of external machine learning resources. To remedy this, new methods are required to enable data owners, such as hospitals, to share their datasets publicly, while preserving both patient privacy and modeling utility. We propose NeuraCrypt 2 , a private encoding scheme based on random deep neural networks. NeuraCrypt encodes raw patient data using a randomly constructed neural network known only to the data-owner, and publishes both the encoded data and associated labels publicly. From a theoretical perspective, we demonstrate that sampling from a sufficiently rich family of encoding functions offers a welldefined and meaningful notion of privacy against a computationally unbounded adversary with full knowledge of the underlying data-distribution. We propose to approximate this family of encoding functions through random deep neural networks. Empirically, we demonstrate the robustness of our encoding to a suite of adversarial attacks and show that NeuraCrypt achieves competitive accuracy to non-private baselines on a variety of x-ray tasks. Moreover, we demonstrate that multiple hospitals, using independent private encoders, can collaborate to train improved x-ray models. Finally, we release a challenge dataset 3 to encourage the development of new attacks on NeuraCrypt. * Equal contribution 2 github.com/yala/NeuraCrypt 3 github.com/yala/NeuraCrypt-Challenge Preprint. Under review.

show abstract

“…The curvature change-based model proposed in the literature also achieves some identification results. The P-M model and its improved model are both second-order PDE models, which produce "step effect" in recognition, while the higher-order PDE model proposed by scholars can effectively solve the problem [5]. The fourth-order PDE model (Y-K model) is proposed in the literature to eliminate the "step effect," but the speckle phenomenon also appears.…”

Section: Related Workmentioning

confidence: 99%

“…In addition, the literature proposes a nonlocal estimation-based image coloring method by combining the texture features of the image. In paper [5], the authors use higher-order regularization for image coloring, which can suppress the color diffusion at the edges of the image. The literature proposes a method for selecting seed pixels that improve the effectiveness of manually adding colors.…”

Section: Related Workmentioning

confidence: 99%

Image Recognition of Pledges of Capital Stock in Small- and Medium-Sized Enterprises Based on Partial Differential Equations

Zhou¹

2021

Advances in Mathematical Physics

View full text Add to dashboard Cite

Image recognition is one of the core research directions in the field of computer vision research, which can be divided into general image recognition and fine-grained image recognition. General image recognition refers to the recognition of different types of objects; fine-grained image recognition refers to the recognition of different subclasses in the same broad class of objects, such as SME financing inventory pledge image recognition. In this paper, we propose a partial differential equation-based image recognition method for SME financing inventory pledges and conduct detailed analysis and experiments. Compared with general images, partial differential equation-based SME financing inventory pledges image recognition is difficult to recognize due to data characteristics such as small differences in features between classes, large differences in features within classes, and a small percentage of targets in the image. To address the problem that existing methods ignore the role of shallow features on fine-grained image recognition, this paper proposes a fine-grained image recognition method based on partial differential equations. By analyzing the important role of shallow features for fine-grained image recognition, a feature fusion method with adaptive weights is proposed. Using this method to fuse shallow and high-level semantic features for recognition, the role of shallow features in fine-grained image recognition is fully exploited. In addition, the proposed method does not change the order of magnitude of the model parameters and is highly transferable. The relevant experimental results verify the effectiveness of the proposed method.

show abstract

Structural Image De-Identification for Privacy-Preserving Deep Learning

Cited by 11 publications

References 23 publications

Analytical and Integrated Framework for Privacy Presrvation of Images in Cloud Ecosystem

Analytical and Integrated Framework for Privacy Presrvation of Images in Cloud Ecosystem

NeuraCrypt: Hiding Private Health Data via Random Neural Networks for Public Training

Image Recognition of Pledges of Capital Stock in Small- and Medium-Sized Enterprises Based on Partial Differential Equations

Contact Info

Product

Resources

About