Faces in places: humans and machines make similar face detection errors.

Area of Science:

• Cognitive Neuroscience
• Computer Vision
• Human Visual Perception

Background:

• Human visual system excels at face detection, sometimes leading to false positives in random patterns.
• Machine vision, notably the Viola-Jones algorithm, has achieved efficient and accurate face detection.

Purpose of the Study:

• To investigate if patterns mistakenly identified as faces by the Viola-Jones algorithm also deceive human observers.
• To compare human and algorithmic face detection error patterns.

Main Methods:

• Stimuli included real faces, illusory faces (algorithm false positives), and non-faces from movies.
• Observers viewed paired stimuli for 20ms and performed a gaze-directed task.
• Manual response tasks were also employed to assess different processing levels.

Main Results:

• Illusory faces were more frequently mistaken for real faces than non-faces.
• Rotation affected real and illusory faces differently, with illusory faces becoming less error-prone when rotated.
• Coloration improved overall performance but did not alter error patterns; manual responses eliminated the illusory face preference.

Conclusions:

• Humans exhibit similar face detection errors to the Viola-Jones algorithm under specific conditions (brief, gaze-directed presentation).
• The relative spatial arrangement of features appears crucial for both human and algorithmic face detection.
• Efficient human face recognition likely involves pre-attentive processing of simple, early visual features.