English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Hand-crafted image quality metrics, such as PSNR and SSIM, are commonly usedto evaluate model privacy risk under reconstruction attacks. Under thesemetrics, reconstructed images that are determined to resemble the original onegenerally indicate more privacy leakage. Images determined as overalldissimilar, on the other hand, indicate higher robustness against attack.However, there is no guarantee that these metrics well reflect human opinions,which, as a judgement for model privacy leakage, are more trustworthy. In thispaper, we comprehensively study the faithfulness of these hand-crafted metricsto human perception of privacy information from the reconstructed images. On 5datasets ranging from natural images, faces, to fine-grained classes, we use 4existing attack methods to reconstruct images from many differentclassification models and, for each reconstructed image, we ask multiple humanannotators to assess whether this image is recognizable. Our studies revealthat the hand-crafted metrics only have a weak correlation with the humanevaluation of privacy leakage and that even these metrics themselves oftencontradict each other. These observations suggest risks of current metrics inthe community. To address this potential risk, we propose a learning-basedmeasure called SemSim to evaluate the Semantic Similarity between the originaland reconstructed images. SemSim is trained with a standard triplet loss, usingan original image as an anchor, one of its recognizable reconstructed images asa positive sample, and an unrecognizable one as a negative. By training onhuman annotations, SemSim exhibits a greater reflection of privacy leakage onthe semantic level. We show that SemSim has a significantly higher correlationwith human judgment compared with existing metrics. Moreover, this strongcorrelation generalizes to unseen datasets, models and attack methods.

Privacy Assessment on Reconstructed Images: Are Existing Evaluation Metrics Faithful to Human Perception?