•  
  •  
 

Abstract

With the improvement of artificial intelligence and deep learning techniques, especially deep convolutional generative adversarial network (DCGAN), there has been a significant development in personal identity and generating images through facial reconstruction systems. This study focuses on proposing a model of personal image reconstruction from forensic sketches using DCGAN. The model comprises two networks: a generator to convert sketch images into real images and a feature network to determine the similarity of the generated images to real ones. Forensic sketches provided by relevant authorities are used as inputs to the proposed model. These sketches include details and information on the perpetrators or missing persons obtained from witnesses or the missing person parents. Prominent facial features extracted from the reconstructed images aid in the process of personal image reconstruction. The proposed model shows good results, achieving up to 99% accuracy in the generated images. The error ratio is reported to be as low as 0.92% based on the evaluation using the CUHKFaces dataset. This study presents a new approach to reconstructing human face images from forensic sketches using DCGAN.

Keywords

DCGAN, Deep learning, Forensic image reconstructing, pix2pix translation, Sketch-to-image

Subject Area

Computer Science

First Page

730

Last Page

739

Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Receive Date

2-8-2024

Revise Date

5-21-2024

Accept Date

5-23-2024

Download

Share

COinS