Non-Global Privacy Protection Facing Sensitive Areas in Face Images
Bibliographic record
Abstract
Face images, as an information carrier, are naturally weak in privacy. If they are collected and analyzed by malicious third parties, personal privacy will leak, and many other unmeasurable losses will occur. Differential privacy protection of face images is mainly being studied under non-interactive frameworks. However, the ε-effect impacts the entire image under these frameworks. Besides, the noise influence is uniform across the protected image, during the realization of the Laplace mechanism. The differential privacy of face images under interactive mechanisms can protect the privacy of different areas to different degrees, but the total error is still constrained by the image size. To solve the problem, this paper proposes a non-global privacy protection method for sensitive areas in face images, known as differential privacy of landmark positioning (DPLP). The proposed algorithm is realized as follows: Firstly, the active shape model (ASM) algorithm was adopted to position the area of each face landmark. If the landmark overlaps a subgraph of the original image, then the subgraph would be taken as a sensitive area. Then, the sensitive area was treated as the seed for regional growth, following the fusion similarity measurement mechanism (FSMM). In our method, the privacy budget is only allocated to the seed; whether any other insensitive area would be protected depends on whether the area exists in a growing region. In addition, when a subgraph meets the criterion for merging with multiple seeds, the most reasonable seed to be merged would be selected by the exponential mechanism. Experimental results show that the DPLP algorithm satisfies ε-differential privacy, its total error does not change with image size, and the noisy image remains highly available.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.006 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.004 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".