University of Twente Student Theses
Exploring identity matching for low quality images with the help of a pipeline for synthetic face generation
Groffen, M.M.L. (2022) Exploring identity matching for low quality images with the help of a pipeline for synthetic face generation.
|
PDF
58MB |
| Abstract: | In this paper we propose a pipeline for synthetic low resolution face generation as an alternative to image downsampling and real-world low resolution datasets for use in forensic cases. The goal of the pipeline is recreating physically accurate low resolution face images in a 3D space. We were able to incorporate a state-of-the-art conditioned machine learning algorithm to generate a realistic synthetic high resolution gallery dataset, by combining StyleGAN2 and attribute based latent space exploration. Using single image 3D reconstruction and a physically based renderer, an identity preserving pipeline was introduced that allows for one-to-one gallery to low resolution probe dataset generation, while enabling flexible pose, lighting, resolution and compression adjustment. Using volumetric path tracing, subsurface light scattering within human skin was emulated. To get further insight into our pipeline output, facial recognition experiments were conducted using state-of-the-art commercial and open-source facial recognition software and super-resolution upscaling using a convolutional neural network. Comparison to the real-world face image dataset SCFace was also conducted to test for potential applicability of our pipeline in forensic cases. Lack of accurate optical aberration and sensor characteristics resulted in a significantly different facial recognition performance on our synthetic dataset, making current application of our pipeline in forensic scenarios unfit. Thorough description of design choices and background make our research however an interesting stepping-stone for future research. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 50 technical science in general, 54 computer science |
| Programme: | Embedded Systems MSc (60331) |
| Link to this item: | https://purl.utwente.nl/essays/94003 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page