Neuroengineering

Vision and Bionic Eye Design

The visual system is the most studied part of the brain. Many low level visual mechanisms have been characterised to the "first order", and higher level, more cognitive, visual processes are also well studied. The visual system has two primary goals. To see and recognize objects in the three-dimensional (3D) spatial environment, and help the individual to navigate around, and interact with, objects within the environment. Object recognition primarily involves the visual system, and may also depend on interactions between memory systems and the visual system. Whereas, navigation depends on the interaction between the visual system and the sensorimotor system, which is in charge of controlling the limbs.

Given that vision is so well studied, it is probably the best sensory modality for which high level cognitive processes can be most realistically simulated by mathematical models. Thus it is probably also the best sensory modality within which one can study neuromodulation of sensory processing, as a means of studying neural processing under different states of consciousness (i.e. asleep vs awake).

Our group has interests in all kinds of aspects of vision including: color vision, depth perception, image segmentation, object recognition, object tracking, motion detection, navigation. We are interested in approaching the modeling of these processes from both neural modeling, and more practical computer vision perspectives.

In addition to basic vision research, through a collaboration with National ICT Australia Victoria Research Laboratory, we are involved in the development of a bionic eye to aid the visually impaired. A primary cause of severe visual impairment is macular degeneration. This occurs when there is damage to the macula part of the retina, which lies at the back of the eye where light is focussed and changed into nerve signals, that are sent to the brain. A bionic eye implant could bypass the diseased cells in the retina and electrically stimulate the remaining viable nerve cells. This would require a computer chip that sits in the back of the individual's eye, linked up to a mini video camera built into glasses that they wear. Images captured by the camera would be transmitted to the chip, which would translate them into electical impulses that the brain can interpret.

Previous Research Projects

Depth Perception: Neural Modeling of 3D Shape-From-Texture

Depth perception is a crucial aspect for both object recognition and the navigation of 3D visual scenes, which are typically modeled using two-dimensional (2D) vision approaches. 2D approaches greatly simplify the visual environment and reduce the dimensionality of factors that can be applied to segmentation of the visual scene. 3D shape-from-texture (SFT), is a form of depth perception, which refers to the perception of 3D shape one experiences when one monocularly views a textured surface. Essentially, light rays reflected from the 3D surface are projected onto the 2D retina of the observer. The texture on the 3D surface is thus projected onto the 2D retina, but it is distorted by the projection. This distortion of the texture can be used as a cue for 3D shape.

The aim of this work was to develop a neural model of how visuo-cortical areas interact to convert a textured 2D image (i.e. the retinal image of the projected surface) into a neural representation of 3D shape. The neural model of SFT provides a physiologically plausible model of visual cortical areas V1, V2 and V4 that can explain the psychophysical SFT data presented in Todd and Akerstrom. 1987, J. Exp. Psych., 13, 242. Two basic problems are solved to achieve this: (1) Patterns of spatially discrete 2D texture elements are grouped into a spatially smooth surface representation of 3D shape. (2) Changes in the statistical properties of texture elements across space induce the perceived 3D shape of this surface representation. This is achieved in the model through multiple-scale filtering of a 2D image, followed by a cooperative-competitive feedback loop that coherently groups texture elements into boundary webs at the appropriate depths using a scale-to-depth map and a subsequent depth competition stage. These boundary webs then gate filling-in of surface lightness signals in order to form a 3D surface percept. In addition to the Todd and Akerstrom data the model can simulate 3D percepts of an elliptical cylinder, a slanted plane, and a photo of a golf ball.

Publications on Vision and Bionic Eye Design

Student projects in Vision and Bionic Eye Design

If you want to learn more about our neural modeling of vision/computer vision research, please contact:
Levin Kuhlmann
email: l.kuhlmann@ee.unimelb.edu.au
+613 8344 6689

Or, if you want to learn more about our bionic eye research, please contact:
Hamish Meffin
email: hmeff@ee.unimelb.edu.au