Human Face Reconstruction using Divine Proportions and Gestalt for Occluded Video Face Recovery in Forensic Analysis using Deep Learning
S. Anita¹, S. Prema²

¹S. Anita, Department of Computer Science, B.M.S. College for Women, Basavangudi, Bangalore (Karnataka), India. 

²Dr. S. Prema, Department of Computer Science, Arulmigu Arthanareeswarar Arts and Science College, Thiruchengodu (Tamil Nadu), India.  

Manuscript received on 30 January 2024 | Revised Manuscript received on 08 February 2024 | Manuscript Accepted on 15 February 2024 | Manuscript published on 28 February 2024 | PP: 1-5 | Volume-11 Issue-2, February 2024 | Retrieval Number: 100.1/ijies.B108911020224 | DOI: 10.35940/ijies.B1089.11020224

Open Access | Editorial and Publishing Policies | Cite | Zenodo | OJS | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: Forensic video analysis has been used in diverse kind of high-profile cases, global discrepancies, and conflict zones. It is a three-phase process involving the scientific examination, comparison, and evaluation of video evidence in legal matters. Human facere construction using deep learning for occluded video face recovery to aid in forensic analysis is the main objective of this paper. Forensic facial reconstruction is a combination of both scientific methods and artistic skill. In this paper, we introduce a method for reconstructing human faces occluded by short noise in nighttime video clips. A skull database is created with unique skull models that vary in shape, form, and proportion. A mathematical model of the human body, utilising the golden ratio algorithm, is proposed and used to determine the proportions of occluded faces. The closure principle of Gestalt theory of visual perception is used to fill in the missing parts of a face design and to create a complete face image using a generative adversarial network (GAN). The proposed model is found to have a 50% lower reduced Median error rate and a 20% lower standard deviation than PrNet, and a 10% lower Mean error rate than 3Dddfav2.

Keywords: Forensic, Face Reconstruction, Golden Ratio, Gestalt.
Scope of the Article: Industrial, Financial and Scientific Applications of All Kinds