The week's pick
Random Articles
Reseach Article
AI-Assisted Criminal Face Generation from Witness Descriptions
| International Journal of Computer Applications |
| Foundation of Computer Science (FCS), NY, USA |
| Volume 187 - Number 104 |
| Year of Publication: 2026 |
| Authors: Ajinkya Valanjoo, Atharva Badhe, Ayush Bohra, Harsh Kotwal, Viresh Warikoo |
10.5120/ijcaf03e5d652293
|
Ajinkya Valanjoo, Atharva Badhe, Ayush Bohra, Harsh Kotwal, Viresh Warikoo . AI-Assisted Criminal Face Generation from Witness Descriptions. International Journal of Computer Applications. 187, 104 ( May 2026), 23-31. DOI=10.5120/ijcaf03e5d652293
Abstract
Criminal investigations in developing nations face a critical issue: The memory of the witness fades rapidly, while there are limited forensic sketch artists. In 2022, over 4,45,000 crimes against women have been reported in India, yet it maintains only 155 police officers per 1,00,000 citizens - well below the UN standard that is 222. So, we present a proof-of-concept system that addresses this gap by integrating modern diffusion models into forensic workflows. Through comparative evaluation of FLUX.1-dev and FLUX.2-klein-4b, we demonstrate that the latter achieves 97% faster generation (2-4 seconds vs. 80-177 seconds on RTX 3060) while reducing VRAM requirements by 30% (8.4GB vs. 12GB). Our implementation generates facial features from witness descriptions in 2 to 4 seconds using consumer hardware, transforming forensic composite generation from a ”coffee break workflow” to truly interactive real time iteration. Our system uses structural similarity matching for database queries. Through qualitative evaluation and deployment testing, we demonstrate that modern generative models can be practically integrated into law enforcement contexts where resources are quite limited. We discuss technical architecture, rationale for model selection, deployment considerations, and legal frameworks specific to India, and identify the key challenges that will be addressed in future work.
References
- National Crime Records Bureau, “Crime in India 2022,” Min-istry of Home Affairs, Govt. of India, 2023.
- C. D. Frowd, P. J. B. Hancock, and D. Carson, “EvoFIT: A holistic, evolutionary facial imaging technique for creating composites,” ACM Trans. Applied Perception, vol. 1, no. 1,pp. 19-39, 2004.
- S. Y. Chen, W. Su, L. Gao, S. Xia, and H. Fu, “DeepFace-Drawing: Deep generation of face images from sketches,” ACM Trans. Graphics, vol. 39, no. 4, pp. 1-16, 2020.
- S. Yu, J. Liu, and K. M. Lam, “Semi-Siamese network for sketch-to-photo retrieval,” in Proc. IEEE CVPR, 2021, pp. 8007-8016.
- J. Wang and L. Zhang, “PI-GAN: A novel generative adver-sarial network for photo-realistic face image synthesis from sketches,” 2022.
- Y. Li, X. Chen, F. Yang, et al., “DeepFacePencil: Creating face images from freehand sketches,” in Proc. ECCV, 2020,pp. 603-620.
- S. Ghosh, A. Hiranandani, O. Kumar, and R. Nachane, “Do generative AI models output harm while representing non-Western cultures,” arXiv:2407.14779, 2024.
- R. Leyva, Y. Wang, and T. Zhu, “Demographic bias effects on face image synthesis,” in Proc. IEEE CVPRW, 2024, pp. 3892-3901.
- L. Zhang, A. Rao, and M. Agrawala, “Adding con-ditional control to text-to-image diffusion models,” arXiv:2302.05543, 2023.
- C. Mou, X. Wang, L. Xie, et al., “T2I-Adapter: Learning adapters to dig out more controllable ability for text-to-image diffusion models,” arXiv:2302.08453, 2023.
- J. Zhang, Y. Liu, and H. Chen, “FluxSchell: High-fidelity multimodal generation with sparse text and visual inputs,” 2024.
- Li, Y. Zhang, and M. Wang, “Bridging the gap between text and face images with contrastive learning,” 2024.
- D. Chen, L. Wang, and X. Liu, “Beyond the sketch: A deep learning framework for text-guided face synthesis,” 2023.
- K. Ka¨rkka¨inen and J. Joo, “FairFace: Face attribute dataset for balanced race, gender, and age,” in Proc. IEEE WACV, 2021, pp. 1548-1558.
- T. Karras, S. Laine, and T. Aila, “A style-based generator ar-chitecture for generative adversarial networks,” in Proc. IEEE CVPR, 2019, pp. 4401-4410.
- L. Song, X. Wu, and Y. Chen, “SketchFace: A large-scale dataset for human sketch-to-face synthesis,” 2024.
- R. Sharma, A. Gupta, and V. Kumar, “PencilSketch-to-face: Challenges and approaches in criminal identification sys-tems,” 2021.
- Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, “Image quality assessment: From error visibility to structural similarity,” IEEE Trans. Image Processing, vol. 13, no. 4, pp. 600-612, 2004.
Index Terms
Keywords