“Ghostly Gaze” Illusion

from Michael’s   Visual Phenomena & Optical Illusions


What to see

The neighbouring face image is obviously defocused (blurred), and the lady looks to the left. Now focus by either pressing the space bar, or using the slider to set blur to zero. Yes, quite sharp now, but not looking to the left any more!

I also notice a hysteresis: when going slowly from focused to defocused, the point where perception switches seeming gaze direction requires more defocus than on the reverse until the gaze flips to the left again.

What to do

To convince yourself that it is really just defocus, you could walk away from your monitor – decreasing size effectively defocuses the image in your eye. Alternatively, you can defocus by screwing up your eyes, taking off glasses or any other method that blurs your vision.

There is also an alternate face to try out.


Dr. Jenkins writes in his paper of 2007: “The figure was constructed by combining spatial-frequency information from two whole-face images that differed only in their gaze direction. A low-pass Gaussian filter (9 cycles per image width) was applied to the right-gaze version of the image, and a high-pass Gaussian filter (13.3 cycles per image width) was applied to the left-gaze version.”

He uses a method, described in 1994 by Philippe G. Schyns & Aude Oliva, that incorporates two slightly different images in one single hybrid image: one image high-pass filtered, the other low-pass filtered. When viewed in focus, the higher spatial frequency image ‘wins’. When blurring, by whatever means, only the low-pass image remains. This technique can be applied to many image types, it is also employed by “Mr. Angry and Mr. Smile” (next page).

This demonstration was created with kind permission of Dr. Rob Jenkins, Department of Psychology, University of Glasgow, UK.


Jenkins R (2007) The lighter side of gaze perception. Perception 36:1266–1268

Schyns PG, Oliva A (1994) From blobs to boundary edges: Evidence for time and spatial scale dependent scene recognition. Psychol Sci 5:195–200