Contents
Often, yes. Deleting EXIF removes the file’s tags, the timestamp, the GPS coordinates and the camera model, but it changes no pixels, and three signals that identify people sit entirely in the pixels: your face, the scene around you, and the fingerprint of the sensor that took the shot. Each survives an EXIF wipe untouched, and each has its own research literature. Stripping metadata removes the easy, exact traces and is worth doing, but on the evidence it does not make a photo of a person anonymous.
What an EXIF wipe does, and does not, do
EXIF, XMP and IPTC are structured text blocks attached to the file, holding the camera model, timestamps and often GPS coordinates. Because they sit apart from the picture, a remover or a re-save clears them without changing a pixel. That is exactly why the wipe has limits: once the tags are gone, anything still identifying has to be read from the unchanged picture, and there is a great deal there to read.
The face
A visible face is the most direct route to a name. Margin-based recognition networks match faces at close to perfect accuracy on clean benchmarks, and ArcFace reaches about 99.8% on LFW (Deng et al., CVPR 2019). The important caveat is quality: Kim and colleagues note that “Recognition in low quality face datasets is challenging because facial attributes are obscured and degraded” (AdaFace, CVPR 2022). So a large, sharp, well-lit face is close to an identifier, while a small or blurry one is far less reliable. Either way, deleting the EXIF changed none of it.
Attribute inference makes the same point from another angle. In a megastudy over 2,646 facial images of 969 individuals across 349 binary attributes, Tkachenko and Jedidi found 82 of them, about 23%, predictable better than random from the image alone (Scientific Reports 2023). The predictable signal is mostly demographic, not personality or competence, and the authors flag “smartphone camera artifacts, BMI, skin properties, and facial hair” as top candidate non-demographic signals, the device and the body imprinting themselves on the picture regardless of any tag.
The scene
The background places you, and none of it is metadata. Image geolocation is a mature field: PlaNet showed a network could attain “superhuman levels of accuracy in some cases” (Weyand, Kostrikov, Philbin, ECCV 2016), GeoCLIP localises images worldwide against a learned, continuous gallery of coordinates (Vivanco Cepeda et al., NeurIPS 2023), and PIGEON lands 40.4% of its predictions within 25 km of the true location, reaching 92.0% country accuracy with a median error of 44.4 km as a separate measure (Haas et al., CVPR 2024). Precision is strong at the country or city scale and weaker at the street, but a storefront, a skyline or a reflection can narrow a location with no GPS tag, which is why deleting the GPS field does not close the location question.
The sensor fingerprint outlives the wipe
The subtlest survivor is the device fingerprint. Every camera imprints a faint, unique noise pattern, its photo-response non-uniformity, or PRNU, onto the pixels it records, introduced for camera source identification by Lukáš, Fridrich and Goljan (IEEE TIFS 2006). Because it lives in the pixels, a metadata wipe does nothing to it, and Goljan and Fridrich report that “The fingerprint survives a wide range of common image processing operations, including lossy compression, filtering, and gamma adjustment” (Proc. SPIE 6819, 2008), describing the pattern as “biometrics for sensors.” The technique scales: Goljan, Fridrich and Filler ran a large-scale test across 6,896 cameras and over a million images (Proc. SPIE 7254, 2009).
It is powerful but conditional. Linking a photo to a specific camera needs that device and clean reference images, and the fingerprint’s reliability is now contested on the latest phones, where Iuliani, Fontani and Piva warn that “with the advent of computational photography, most recent devices heavily process the acquired pixels, possibly introducing non-unique artifacts that may reduce PRNU noise’s distinctiveness” (IEEE Access 2021). Nothing about removing EXIF affects it either way. What the fingerprint can and cannot trace is the subject of can a photo be traced without metadata.
Uploading is not anonymising
Posting a photo does not clean it. Most platforms strip EXIF on upload, but they re-encode the visible image rather than rebuild it, and that re-encode is exactly the lossy compression the sensor fingerprint is documented to survive (Goljan and Fridrich, Proc. SPIE 6819, 2008). The face and the scene are left untouched as well, and the platform often keeps the original file you sent. An uploaded, EXIF-free photo has lost its tags and kept every pixel-level identifier it started with.
What removing EXIF actually buys
On the evidence, an EXIF wipe clears the precise machine-readable traces, the GPS pin, the timestamp and the device string, and nothing more. The three signals that actually identify people, the face, the scene and the sensor fingerprint, all sit in the pixels and pass through untouched. None of this makes identification certain: coarse geolocation is stronger than street-level, low-quality faces resist matching, and PRNU needs the candidate device in hand. But the answer to the title is that removing EXIF does not, by itself, make you unidentifiable. Treat it as one layer of reduction among several, not a finish line. For the wider map of what survives a scrub, see does removing metadata make a photo anonymous.
Sources
- Lukáš, Fridrich, Goljan (2006). Digital Camera Identification from Sensor Pattern Noise. IEEE TIFS 1(2):205-214.
- Goljan, Fridrich (2008). Camera Identification from Cropped and Scaled Images. Proc. SPIE 6819, Security, Forensics, Steganography, and Watermarking of Multimedia Contents X.
- Goljan, Fridrich, Filler (2009). Large Scale Test of Sensor Fingerprint Camera Identification. Proc. SPIE 7254.
- Iuliani, Fontani, Piva (2021). A leak in PRNU based source identification. Questioning fingerprint uniqueness. IEEE Access 9.
- Weyand, Kostrikov, Philbin (2016). PlaNet: Photo Geolocation with Convolutional Neural Networks. ECCV 2016.
- Haas, Skreta, Alberti, Finn (2024). PIGEON: Predicting Image Geolocations. CVPR 2024:12893-12902.
- Vivanco Cepeda, Nayak, Shah (2023). GeoCLIP: Clip-Inspired Alignment between Locations and Images for Effective Worldwide Geo-localization. NeurIPS 2023.
- Deng, Guo, Xue, Zafeiriou (2019). ArcFace: Additive Angular Margin Loss for Deep Face Recognition. CVPR 2019:4690-4699.
- Kim, Jain, Liu (2022). AdaFace: Quality Adaptive Margin for Face Recognition. CVPR 2022.
- Tkachenko, Jedidi (2023). A megastudy on the predictability of personal information from facial images. Scientific Reports 13:21073.