Method for forensic identification by automatically comparing a 3D model of a skull with one or more photos of a face

Patent metadata

Field	Value
Publication number	US-12536742-B2
Application number	US-202218575992-A
Country	US
Kind code	B2
Filing date	Jun 29, 2022
Priority date	Jul 2, 2021
Publication date	Jan 27, 2026
Grant date	Jan 27, 2026

Abstract

Disclosed is a method for identifying multiple skulls and different subjects, using (i) a 3D model of each skull and (ii) one or more clear photos of the face of each candidate subject, wherein the method comprises the steps of: detecting craniometric points on the 3D skull models; estimating the thickness and direction of the soft facial tissue; detecting cephalometric landmarks in the face photos; and filtering on the basis of morphometric consistencies between the skull and the face, craniofacial superimposition in multiple photos, and decision making.

First claim

The invention claimed is:

1 . A method for identification of multiple skulls with different subjects, when (i) a three-dimensional model of each skull and (ii) one or more indubitable facial photographs of each of the candidate subjects are available, the method comprising the following steps:
detecting craniometric landmarks on the three-dimensional model of each skull by:
applying a template where the craniometric landmarks have been located;
adapting the dimensions of said template to the three-dimensional model of the skull;
aligning the template with said model in such a way that both are overlaid,
transferring the craniometric landmarks to the three-dimensional model of the skull;
refining the position of the craniometric landmarks according to specific criteria for each landmark using information on symmetry axes, nostrils and eye socket contours, end landmarks of the anatomical axes, position of sutures;
automatically estimating facial soft tissue thickness and direction, by:
simultaneously estimating thickness in mm and directional vector;
using a machine learning model previously trained with real data of soft tissue, lineage, age, sex of the subject and BMI;
detecting cephalometric landmarks in facial photographs by:
using a machine learning model pre-trained with thousands of photographs and their respective cephalometric landmarks
automatic craniofacial superimposition with one or multiple photographs at the same time, by:
using a priori information: focal, pose and camera-subject distance estimation;
modeling of the cephalometric landmark marking error and of the uncertainty of the soft tissue thickness and direction;
estimating soft tissue shared between photographs in cases where several photographs of the same subject are available;
automatic decision-making step, aggregating information from:
the quality of the available photographs
the morphometric consistency between the skull and the face
the average overlay error on the photographs of each of the candidate subjects
the plausibility of the soft tissue with respect to the human population.

2 . The method according to claim 1 comprising, before the automatic craniofacial superimposition step, an automatic step of estimating the camera-subject distance of the one of more facial photographs, comprising:
using photographs in which the face shows any pose, obtained with any focal length;
using a machine learning model previously trained with thousands of photographs whose camera-subject distance is known.

3 . The method according to claim 1 comprising, before the automatic craniofacial superimposition step, an automatic step of discarding candidates whose skull is not compatible with the photograph, the steps of:
estimating 3D cranial indices from 2D facial indices; and
using a machine learning model previously trained with thousands of photographs and 3D skull models.

4 . The method according to claim 2 comprising, before the automatic craniofacial superimposition step, an automatic step of discarding candidates whose skull is not compatible with the photograph, the steps of:
estimating 3D cranial indices from 2D facial indices; and
using a machine learning model previously trained with thousands of photographs and 3D skull models.

Assignees

• Panacea Cooperative Res S Coop
• Univ Granada

Inventors

• Ibáñez Panizo Óscar
• Valsecchi Andrea
• Bermejo Nievas Enrique
• Martos Fernández Rubén
• Mesejo Santiago Pablo
• Martínez Moreno Práxedes
• Gómez Trenado Guillermo
• Cordón García Óscar
• Damas Arroyo Sergio
• Guerra Martí Rosario
• Alemán Aguilera Inmaculada

Classifications

• G06V2201/07

  Target detection · CPC title

• G06T2207/20221

  Image fusion; Image merging · CPC title

• G06V40/168

  Feature extraction; Face representation · CPC title

• G06V10/24

  Aligning, centring, orientation detection or correction of the image · CPC title

• G06T7/70

  Determining position or orientation of objects or cameras (camera calibration G06T7/80) · CPC title