Related Experiment Videos
Effects of high-pass and low-pass spatial filtering on face identification
N P Costen1, D M Parker, I Craw
1University of Aberdeen, Scotland. costen@hip.atr.co.jp
Perception & Psychophysics
|May 1, 1996
Summary
Face recognition accuracy declines non-linearly with reduced spatial frequencies, indicating a critical range for identification. This suggests specific spatial frequency bands are essential for recognizing faces.
Area of Science:
- Cognitive Science
- Neuroscience
- Computer Vision
Background:
- Human face recognition is a complex cognitive process.
- Understanding the visual information critical for face identification is crucial for developing robust recognition systems.
Purpose of the Study:
- To investigate the role of spatial frequencies in face recognition.
- To determine the specific range of spatial frequencies essential for accurate face identification.
Main Methods:
- Face images were degraded using block averaging and Fourier filtering (low-pass and high-pass).
- Recognition accuracy and response times were measured across varying spatial-frequency ranges.
- Experiments controlled for image contrast variations.
Main Results:
- Recognition accuracy declined non-linearly as the spatial-frequency range was reduced.
- Performance degradation was more rapid with quantized and high-pass filtered images.
- A critical band of 8-16 cycles per face was identified as preferential for face identification.
Conclusions:
- Face recognition relies on a specific band of spatial frequencies, not just contrast or line information.
- These findings inform current models of human face identification and computer vision algorithms.