Physics Maths Engineering
Youchen Fan,
Mingyu Qin,
Huichao Guo
Huichao Guo
Department of Electronic and Optical Engineering, Space Engineering University
Peer Reviewed
The range-gated laser imaging instrument can capture face images in a dark environment, which provides a new idea for long-distance face recognition at night. However, the laser image has low contrast, low SNR and no color information, which affects observation and recognition. Therefore, it becomes important to convert laser images into visible images and then identify them. For image translation, we propose a laser-visible face image translation model combined with spectral normalization (SN-CycleGAN). We add spectral normalization layers to the discriminator to solve the problem of low image translation quality caused by the difficulty of training the generative adversarial network. The content reconstruction loss function based on the Y channel is added to reduce the error mapping. The face generated by the improved model on the self-built laser-visible face image dataset has better visual quality, which reduces the error mapping and basically retains the structural features of the target compared with other models. The FID value of evaluation index is 36.845, which is 16.902, 13.781, 10.056, 57.722, 62.598 and 0.761 lower than the CycleGAN, Pix2Pix, UNIT, UGATIT, StarGAN and DCLGAN models, respectively. For the face recognition of translated images, we propose a laser-visible face recognition model based on feature retention. The shallow feature maps with identity information are directly connected to the decoder to solve the problem of identity information loss in network transmission. The domain loss function based on triplet loss is added to constrain the style between domains. We use pre-trained FaceNet to recognize generated visible face images and obtain the recognition accuracy of Rank-1. The recognition accuracy of the images generated by the improved model reaches 76.9%, which is greatly improved compared with the above models and 19.2% higher than that of laser face recognition.
Range-gated laser imaging captures face images in dark environments, enabling long-distance face recognition at night. However, laser images often have low contrast, low signal-to-noise ratio (SNR), and no color, making them challenging to use directly for recognition.
Laser images lack color and detail, which makes observation and recognition difficult. Converting them to visible images improves visual quality and makes it easier to identify faces, especially in low-light conditions.
SN-CycleGAN is a laser-to-visible image translation model that uses spectral normalization (SN) to stabilize training and improve image quality. It also includes a Y-channel-based content reconstruction loss to reduce errors and preserve structural features.
SN-CycleGAN outperforms models like CycleGAN, Pix2Pix, and StarGAN, achieving a lower Fréchet Inception Distance (FID) score of 36.845. This indicates better visual quality and fewer errors in the translated images.
Spectral normalization stabilizes the training of the generative adversarial network (GAN), preventing issues like mode collapse and improving the quality of the translated images.
The proposed face recognition model retains identity information by directly connecting shallow feature maps to the decoder. It also uses a domain loss function based on triplet loss to maintain style consistency between domains.
The improved model achieves a Rank-1 recognition accuracy of 76.9%, which is 19.2% higher than direct laser face recognition and significantly better than other models like CycleGAN and Pix2Pix.
The Y-channel in the YCbCr color space represents luminance, which is critical for preserving structural details in images. Using it in the loss function helps reduce error mapping and improve translation quality.
This technology is useful for security and surveillance, especially in low-light or nighttime scenarios. It enables accurate face recognition in challenging conditions, such as long-distance monitoring or dark environments.
By converting low-quality laser images into high-quality visible images and using a feature-retention recognition model, the study significantly improves face recognition accuracy in low-light environments.
The study used a self-built dataset of laser-visible face images to train and evaluate the models. This dataset was essential for testing the performance of the proposed SN-CycleGAN and face recognition models.
This technology can be applied in security systems, law enforcement, and surveillance, particularly for nighttime monitoring or in environments where traditional cameras struggle to capture clear images.
Show by month | Manuscript | Video Summary |
---|---|---|
2025 April | 2 | 2 |
2025 March | 80 | 80 |
2025 February | 48 | 48 |
2025 January | 48 | 48 |
2024 December | 54 | 54 |
2024 November | 61 | 61 |
2024 October | 54 | 54 |
2024 September | 52 | 52 |
2024 August | 38 | 38 |
2024 July | 37 | 37 |
2024 June | 29 | 29 |
2024 May | 31 | 31 |
2024 April | 28 | 28 |
2024 March | 8 | 8 |
Total | 570 | 570 |
Show by month | Manuscript | Video Summary |
---|---|---|
2025 April | 2 | 2 |
2025 March | 80 | 80 |
2025 February | 48 | 48 |
2025 January | 48 | 48 |
2024 December | 54 | 54 |
2024 November | 61 | 61 |
2024 October | 54 | 54 |
2024 September | 52 | 52 |
2024 August | 38 | 38 |
2024 July | 37 | 37 |
2024 June | 29 | 29 |
2024 May | 31 | 31 |
2024 April | 28 | 28 |
2024 March | 8 | 8 |
Total | 570 | 570 |