# Cross-Domain Identity Representation for Skull to Face Matching with Benchmark DataSet

arXiv: https://arxiv.org/abs/2507.08329 · HTML: https://mlsanity.com/paper/2507.08329/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- Builds on 3 of its 25 references that are in the map.
- First posted on arXiv 11 Jul 2025.

## Builds on (most cited first, 3 of 3)

- [Learning Representations by Maximizing Mutual Information Across Views](https://mlsanity.com/paper/1906.00910/) (2019)
- [STIR: Siamese Transformer for Image Retrieval Postprocessing](https://mlsanity.com/paper/2304.13393/) (2023)
- [On Matching Skulls to Digital Face Images: A Preliminary Approach](https://mlsanity.com/paper/1710.02866/) (2017)

## Built on by (most cited first, 4 of 4)

- [FCR: Investigating Generative AI models for Forensic Craniofacial Reconstruction](https://mlsanity.com/paper/2508.18031/) (2025)
- [Cranio-ID: Graph-Based Craniofacial Identification via Automatic Landmark Annotation in 2D Multi-View X-rays](https://mlsanity.com/paper/2511.14411/) (2025)
- [SPOT-Face: Forensic Face Identification using Attention Guided Optimal Transport](https://mlsanity.com/paper/2601.09229/) (2026)
- [Cranio-Diff: Diffusion-based Cross-domain Craniofacial Reconstruction with 2D X-ray Skull Guidance and Structural Identity Constraints](https://mlsanity.com/paper/2606.09699/) (2026)

## Abstract (arXiv)

Craniofacial reconstruction in forensic science is crucial for the identification of the victims of crimes and disasters. The objective is to map a given skull to its corresponding face in a corpus of faces with known identities using recent advancements in computer vision, such as deep learning. In this paper, we presented a framework for the identification of a person given the X-ray image of a skull using convolutional Siamese networks for cross-domain identity representation. Siamese networks are twin networks that share the same architecture and can be trained to discover a feature space where nearby observations that are similar are grouped and dissimilar observations are moved apart. To do this, the network is exposed to two sets of comparable and different data. The Euclidean distance is then minimized between similar pairs and maximized between dissimilar ones. Since getting pairs of skull and face images are difficult, we prepared our own dataset of 40 volunteers whose front and side skull X-ray images and optical face images were collected. Experiments were conducted on the collected cross-domain dataset to train and validate the Siamese networks. The experimental results provide satisfactory results on the identification of a person from the given skull.