Image processing effects on the deep face recognition system.

Area of Science:

• Forensic Science
• Computer Vision
• Image Processing

Background:

• Face recognition is crucial for forensic identification, but image quality affects performance.
• Previous studies often used synthetic noise and traditional algorithms, not real-world data or deep learning.

Purpose of the Study:

• To analyze the impact of denoising and enhancement on deep face recognition performance using real forensic case images.
• To compare the effectiveness of different image processing techniques for improving face recognition in practical forensic applications.

Main Methods:

• Collected face images from 33 real forensic identification cases.
• Applied Gaussian filtering (denoising) and a self-snake model (enhancement).
• Evaluated performance using a deep convolutional neural network (MXNet) face recognition system.

Main Results:

• Face image quality significantly influences face recognition system accuracy.
• Image processing techniques, particularly denoising, can enhance image quality and recognition precision.
• Gaussian filtering demonstrated superior results compared to the self-snake model for deep face recognition.

Conclusions:

• Image quality is a critical factor in the performance of deep face recognition systems for forensic identification.
• Denoising methods are more effective than enhancement techniques for improving face recognition accuracy in practical forensic scenarios.
• Gaussian filtering is a suitable preprocessing step for enhancing deep face recognition performance in real-world forensic image analysis.