Psych 129 - Sensory processes

Visual information processing

Receptive fields

• The responses of photoreceptors are collected and summated together by a network of neurons within the retina. This network is composed of four main types of neurons: horizontal cells, bipolar cells, amacrine cells, and ganglion cells.
  
  
• Ganglion cells are the output cells of the retina, and it is their axons bundled together that form the optic nerve.
  
  
• Each ganglion cell has a receptive field on the retina, which is defined by the contiguous region of photoreceptors from which a ganglion cell receives input. Note that ganglion cells do not receive direct input from photoreceptors however; the connection between photoreceptors and ganglion cells is mediated by bipolar cells and other cells in the retinal network.
  
  
• The receptive field of each ganglion cell has a characteristic center-surround property. That is, one portion of the receptive field will be excitatory, and the other inhibitory. These regions are organized in a circularly symmetric fashion so that either the excitatory region is surrounded by the inhibitory region, or vice-versa. Half of the retinal ganglion cells have ON-center/OFF-surround receptive fields, meaning an excitatory center and inhibitory surround, and the other half have OFF-center/ON-surround receptive fields, meaning an inhibitory center and excitatory surround.
  
  
• There are two major types of retinal ganglion cell - parvo-cells and magno-cells - that form two major parallel processing streams through the optic nerve, LGN, and into the visual cortex. The magno-cellular stream has high sensitivity, low spatial resolution, high temporal resolution (quick response), and little or no color selectivity. The parvo-cellular stream has the exact opposite characteristics: low sensitivity, high spatial resolution (approx. 9 times the sampling density of the magno-cells), low temporal resolution (sluggish response), and color selectivity.
  
  
  
  

• Retinal ganglion cells perform an early stage of information processing on the image. That is, rather than sending information about individual pixel values in the retina, these neurons send information about pixel differences. Thus, activity among the array of retinal ganglion cells will be enhanced in those regions of the image containing luminance discontinuities, or places where the luminance changes abruptly across space. Such a representation may be useful for signaling the presence of edges in a scene.
  
  
• The effect of this enhancement of luminance discontinuities may be experienced in the form of Mach bands. When you view a series of adjacent gray patches that differ slightly in their gray-level, the differences in their gray-levels appears to be enhanced at the boundary, even though it is not. This illusion is thought to be produced by retinal processing, although it may be due to later stages of cortical processing as well.

  ---

• There are about one million ganglion cells in each retina. Thus, the net convergence ratio of photoreceptors to ganglion cells is about 100:1.
  
  
• The convergence ratio is not constant across the retina. In the fovea, the convergence ratio is about 1:1, meaning there is about one ganglion cell for each cone in the fovea. As one moves away from the fovea, retinal ganglion cell spacing increases linearly with eccentricity.
  

• This non-uniform packing of ganglion cells in the retina is the main factor that gives rise to high-resolution in the fovea and low-resolution vision in the periphery. This packing arrangement represents a tradeoff between resolution and field-of-view. If the one million ganglion cells were packed uniformly across the entire retina, then one would have only the equivalent of a 1000x1000 image of the entire visual field - which is equivalent to the resolution of an average computer monitor across its 12” screen (not very good). Alternatively, if the same number of ganglion cells were packed uniformly across a much smaller region in order to obtain higher resolution, then one would have tunnel vision. The human visual system (and that of most mammals) attempts to obtain the best of both worlds by changing the sample spacing as a function of eccentricity.
  
  
  

• Nearly all organisms that ‘see’ possess some way of forming an image by bending light and detecting light at different points of the image with photoreceptors. There are many variations on this basic theme.
  
  
• The compound eye of the fly contains many small lenses, each of which forms a miniature eye in itself with 7 photoreceptors. Each such miniature eye is called an ommatidium. The compound eye of the fly contains about 10,000 ommatidia.
  
  
• The fly has poor resolution but good field of view. It is quite good at perceiving motion and using optic flow to guide the control of flight behavior, but it can perceive most objects only as blurry blobs.
  
  
• The jumping spider by contrast has eight eyes, each of which is a lens/camera type eye such as our own. Most of these eyes have fish-eye like lenses, giving them a wide field of view but blurry vision. The two forward looking eyes, though, have very high resolution - almost as good as a cat.
  
  
• The jumping spider uses its low-resolution, wide field-of view eyes to detect the presence of moving objects and prey, and then it quickly reorients its body to image the object of interest with its high-resolution eyes for pattern analysis. The jumping spider is capable of discriminating its prey (flies) from mates (other spiders) using vision alone.
  
  
• The human visual system gets the best of both worlds in one eye - high resolution and wide field of view - by packing the ganglion cells so as to form a non-uniform sampling lattice.