Analyzing images' privacy for the modern web

Squicciarini, Anna; Caragea, Cornelia; Balakavi, Rahul

doi:10.1145/2631775.2631803

Cited by 30 publications

(29 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Image tags provide relevant cues for privacy-aware image retrieval [Zerr et al 2012b] and can become an essential tool for surfacing the hidden content of the deep Web without exposing sensitive details. Additionally, previous works showed that user tags performed better or on par compared with visual features [Squicciarini et al 2014;Tonge and Caragea 2015Zerr et al 2012b]. For example, in our previous work [Tonge and Caragea 2015, we showed that the combination of user tags and deep tags derived from AlexNet performs comparably to the AlexNet based visual features.…”

Section: Best Performing Visual Features Vs Tag Featuresmentioning

confidence: 60%

“…Prior works on privacy prediction [Squicciarini et al 2014[Squicciarini et al , 2017bCaragea 2015, 2016;Zerr et al 2012b] found that the tags associated with images are indicative of their sensitive content. Tags are also crucial for image-related applications such as indexing, sharing, searching, content detection and social discovery [Bischoff Object Since not all images on social networking sites have user tags or the set of user tags is very sparse [Sundaram et al 2012], we use an automatic technique to annotate images with tags based on their visual content as described in our previous work Caragea 2015, 2016].…”

Section: Image Tags (Bag-of-tags Model)mentioning

confidence: 99%

“…We also consider classifiers trained on the best performing features between SIFT, GIST, and their combination as our baselines. Our choice of these features is motivated by their good performance over other visual features such as colors, patterns, and edge directions in prior works [Squicciarini et al 2014;Zerr et al 2012b]. For SIFT, we construct a vocabulary of 128 visual words for our experiments.…”

Section: Sift and Gistmentioning

confidence: 99%

See 2 more Smart Citations

Image Privacy Prediction Using Deep Neural Networks

Tonge

Caragea

2020

ACM Trans. Web

Self Cite

View full text Add to dashboard Cite

Images today are increasingly shared online on social networking sites such as Facebook, Flickr, Foursquare, and Instagram. Image sharing occurs not only within a group of friends but also more and more outside a user's social circles for purposes of social discovery. Despite that current social networking sites allow users to change their privacy preferences, this is often a cumbersome task for the vast majority of users on the Web, who face difficulties in assigning and managing privacy settings. When these privacy settings are used inappropriately, online image sharing can potentially lead to unwanted disclosures and privacy violations. Thus, automatically predicting images' privacy to warn users about private or sensitive content before uploading these images on social networking sites has become a necessity in our current interconnected world.In this paper, we explore learning models to automatically predict appropriate images' privacy as private or public using carefully identified image-specific features. We study deep visual semantic features that are derived from various layers of Convolutional Neural Networks (CNNs) as well as textual features such as user tags and deep tags generated from deep CNNs. Particularly, we extract deep (visual and tag) features from four pre-trained CNN architectures for object recognition, i.e., AlexNet, GoogLeNet, VGG-16, and ResNet, and compare their performance for image privacy prediction. Among all four networks, we observe that ResNet produces the best feature representations for this task. We also fine-tune the pre-trained CNN architectures on our privacy dataset and compare their performance with the models trained on pre-trained features. The results show that even though the overall performance obtained using the fine-tuned networks is comparable to that of pre-trained networks, the fine-tuned networks provide an improved performance for the private class as compared to models trained on the pre-trained features. Results of our experiments on a Flickr dataset of over thirty thousand images show that the learning models trained on features extracted from ResNet outperform the state-of-the-art models for image privacy prediction. We further investigate the combination of user tags and deep tags derived from CNN architectures using two settings: (1) SVM on the bag-of-tags features; and (2) text-based CNN. We compare these models with the models trained on ResNet visual features obtained for privacy prediction. Our results show that even though the models trained

show abstract

Section: Best Performing Visual Features Vs Tag Featuresmentioning

confidence: 60%

Section: Image Tags (Bag-of-tags Model)mentioning

confidence: 99%

Section: Sift and Gistmentioning

confidence: 99%

See 1 more Smart Citation

Image Privacy Prediction Using Deep Neural Networks

Tonge

Caragea

2020

ACM Trans. Web

Self Cite

View full text Add to dashboard Cite

show abstract

“…Authors considered image tags and visual features such as color histograms, faces, edge-direction coherence, and SIFT for the privacy classi cation task. Squicciarini et al [2014 found that SIFT and image tags work best for predicting sensitivity of user's images. Given the recent success of CNNs, Tran et al [2016], and Tonge and Caragea [2016,2018] showed promising privacy predictions compared with visual features such as SIFT and GIST.…”

Section: Related Workmentioning

confidence: 99%

“…Motivated by the fact that increasingly online users' privacy is routinely compromised by using social and content sharing applications [58], recently, researchers started to explore machine learning and deep learning models to automatically identify private or sensitive content in images [35,45,[49][50][51][52]57]. Starting from the premise that the objects and scene contexts present in images impact images' privacy, many of these studies used objects, scenes, and user tags, or their combination (i.e., feature-level or decision-level fusion) to infer adequate privacy classi cation for online images.…”

Section: Introductionmentioning

confidence: 99%

Dynamic Deep Multi-modal Fusion for Image Privacy Prediction

Tonge

Caragea

2019

The World Wide Web Conference

Self Cite

View full text Add to dashboard Cite

With millions of images that are shared online on social networking sites, e ective methods for image privacy prediction are highly needed. In this paper, we propose an approach for fusing object, scene context, and image tags modalities derived from convolutional neural networks for accurately predicting the privacy of images shared online. Speci cally, our approach identi es the set of most competent modalities on the y, according to each new target image whose privacy has to be predicted. e approach considers three stages to predict the privacy of a target image, wherein we rst identify the neighborhood images that are visually similar and/or have similar sensitive content as the target image. en, we estimate the competence of the modalities based on the neighborhood images. Finally, we fuse the decisions of the most competent modalities and predict the privacy label for the target image. Experimental results show that our approach predicts the sensitive (or private) content more accurately than the models trained on individual modalities (object, scene, and tags) and prior privacy prediction works. Also, our approach outperforms strong baselines, that train meta-classi ers to obtain an optimal combination of modalities. CCS CONCEPTS•Security and privacy → Social network security and privacy; KEYWORDS Image privacy prediction; fusion of modalities; decision-level fusion ACM Reference format:

show abstract

Image Pixelization with Differential Privacy

Fan

2018

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Ubiquitous surveillance cameras and personal devices have given rise to the vast generation of image data. While sharing the image data can benefit various applications, including intelligent transportation systems and social science research, those images may capture sensitive individual information, such as license plates, identities, etc. Existing image privacy preservation techniques adopt deterministic obfuscation, e.g., pixelization, which can lead to re-identification with well-trained neural networks. In this study, we propose sharing pixelized images with rigorous privacy guarantees. We extend the standard differential privacy notion to image data, which protects individuals, objects, or their features. Empirical evaluation with real-world datasets demonstrates the utility and efficiency of our method; despite its simplicity, our method is shown to effectively reduce the success rate of re-identification attacks.

show abstract

Analyzing images' privacy for the modern web

Cited by 30 publications

References 46 publications

Image Privacy Prediction Using Deep Neural Networks

Image Privacy Prediction Using Deep Neural Networks

Dynamic Deep Multi-modal Fusion for Image Privacy Prediction

Image Pixelization with Differential Privacy

Contact Info

Product

Resources

About