Though much is known about how binocular neurons in the primary visual cortex respond to stereo imagery, there has yet to be a consensus on how these responses are actually used to compute stereo disparity, the difference in the position of an object between the right image and left image in a stereo pair. We describe a new theory for neural stereo disparity computation using a reformulation of the well-known binocular energy model as an energy response of complex continuous wavelets. These wavelets are used to detect disparity phase interference (DPI), local sinusoidal patterns created in the frequency spectrum when a pair of stereo images are added together. The magnitude of disparity can be approximated mathematically from the frequency of the DPI. Once the disparity is determined for each object, the 3D localization of the object can occur. Describing the binocular complex cell responses with the wavelet transform offers a powerful means of analyzing information content in images, and is highly amenable to the detection of DPI. We believe that this technique represents a promising step towards biomimetic stereo vision.
展开▼
