|M.Sc Student||Shvartz Ishai|
|Subject||The Effect of Noise Patterns on Depth Perception in|
|Department||Department of Electrical Engineering||Supervisors||Professor Alfred Bruckstein|
|Dr. Tamar Avraham|
|Full Thesis text|
This work tests experimentally the ease of perceiving depth in autostereograms. An autostereogram is a single image that contains depth information encoded in repeating vertical strips of a basic 2D-pattern with disparity shifts that effectively describe the three-dimensional information. In designing an autostereogram, one has the freedom to choose the basic pattern, and the repetition frequency, and hence the question of which 2D patterns are best perceived arises naturally. In this work we confirm a theory proposed to explain the process of autostereographic 3D perception, providing a measure for the ease of "lock-in" to the depth information based on the spectral properties of the 2D basic pattern used to produce it. We performed three sets of psychophysical experiments using autostereograms constructed from random 2D noise with different statistics: white, a spectrum of 1/f^(1/2), pink (1/f), a spectrum of 1/f^(3/2) and brown (1/f^2). The experiments tested the ability of human subjects to identify smooth surfaces and to recognize small objects like letters, and searched for the JND (Just Noticeable Difference) in identifying small objects as a function of their size and depth. We can identify a significant advantage of the 1/f noise patterns for fast depth lock-in and fine detail detection, proving that such patterns are optimal choices for autostereogram design.
This work also demonstrates the fact that our stereo-vision mechanism is adapted to work best with natural images, since natural scenery images often exhibit 1/f-spectra.