Processing in the human visual system follows a hierarchy that starts with photore- ceptors and increases in neural complexity with convergence occurring at every stage. The retina is composed of layers containing different types of cell: photoreceptors, horizontal cells, bipolar cells, amacrine cells and retinal ganglion cells. Melanin is a black pigment required in the retina to prevent reflection within the eye and to restore photo-receptor sensitivity after bleaching. For the melanin to perform this function it must be close to the photoreceptors, which in turn must be connected to the nerve cells that process the signals. This necessitates a front-illuminated design, as shown in Figure 3.1. At the centre of the retina is the fovea, which forms a 0.5 mm shallow pit in which light impinges on the photoreceptors directly in the absence of other retinal cells. The are about 104cone cells of
1 - 4 µm diameter in the fovea that are densely packed (approximately 175 x 104 mm−2), giving the highest spatial sampling rate in the retina and hence the greatest acuity.
The two types of photoreceptor cell in the retina are called rods and cones. Rods are sensitive in low luminance (scotopic; 10−6− 10−2 cd/m2, peak sensitivity at around 555
nm) conditions and are bleached in day light whereas cones are sensitive in higher luminance (mesopic; 10−2− 1 cd/m2and photopic; 1 − 106cd/m2) conditions. There are three types of
cone cell that are labelled based on their spectral sensitivity, which is either long wavelength (L-cone; peaking at 575 nm), medium wavelength (M-cone; peaking at 535 nm) or short wavelength (S-cone; peaking at 445 nm) sensitive. Although the three classes have cones
Figure. 3.1: A schematic of the hierarchy of visual processing in the retina showing the light passing through the retinal cells to the photoreceptors. The connections then follow a route from the photoreceptors to the bipolar cells then to the retinal ganglion cells and out via the optic nerve. The horizontal and amacrine cells modulate the retinal receptive field sizes and configurations.(Frisby, 1980)
Figure. 3.2: Distribution of rods and cones in a horizontal line across the retina (from Osterberg, 1935).
have different spectral sensitivities, the cone signal does not indicate the wavelength of light by itself. Colour is determined by a comparison of intensity signals from two types of cone. Comparing signals from L- and M- cones mediates red-green discrimination. Blue-yellow discrimination is mediated by a comparison of S-cone signals with a combination of L- and M- cone signals. The distribution of rods and cones is not uniform across the retina, as shown by Figure 3.2 and there are no rods in the centre of the fovea.
Signals from a selection of nearby photoreceptors are combined in a bipolar cell in a
Figure. 3.3: The centre-surround receptive field. In this diagram ‘−’ represents hyperpo- larisation caused by incident light and ‘+’ represents depolarisation due to incident light. On the left is a ‘on-centre’ receptive field with a depolarising centre and on the right is an ‘off-centre’ receptive field with a hyperpolarising centre.
cells. Ganglion cells fire action potentials, which are discrete events that only occur when the cell’s potential reaches a certain threshold, and the firing rate represents the strength of the cell’s response. The receptive fields of the most common types of ganglion cell (midget and parasol) can be described as either ‘on-centre’ (illumination in the centre depolarises the cell) or ‘off-centre’ (illumination in the centre hyperpolarises the cell). This kind of organisation is called spatial antagonism and is important in visual processing as it takes account of the distribution of light and hence the local contrast. An on-centre receptive field under uniform illumination experiences depolarisation in the centre and hyperpolarisation in the surround, causing a net change in potential (from its resting state) of zero and the cell does not fire (in reality it is never actually zero, as explained in section 3.3). However, if the centre is illuminated (so depolarises) but the surround is not illuminated (so does not hyperpolarise) the net change in potential is positive and the cell fires rapidly. This centre- surround configuration is analogous to a kernel in image processing, which is used as method of detecting features. It also equates to image compression since the signals from many pho- toreceptors are represented in a meaningful way by a single response. The centre of the receptive field follows a direct path from photoreceptors to bipolar cells and the surround follows an indirect path using the lateral connections of the horizontal cells. In the fovea photoreceptors synapse to bipolar cells in a one-to-one manner using only the direct path (the centre), whereas processing in the peripheral retina uses both the direct and indirect paths. This maintains a high rate of spatial sampling in the fovea where acuity is highest. In the peripheral retina receptive field centres are larger indicating a further reduction in sampling rate and hence acuity. Ganglion cell receptive fields tend to be circularly symmet- ric, though in periphery they may become elongated. The centre-surround configuration of ganglion cell receptive fields is additionally modulated by amacrine cells. There are many different types of ganglion cell, each having their own characteristic wiring arrangement, and believed to be specialised for extracting particular types of visual information (Dacey, Peterson, Robinson, & Gamlin, 2003). Recent work has identified a melanopsin-containing ganglion cell that is intrinsically photosensitive (Berson, Dunn, & Takao, 2002). These cells
depolarise in response to light even in the absence of rod and cone inputs. They are thought to provide the primary photoreceptor input to the suprachiasmatic nucleus of the hypotha- lamus, which is responsible for circadian rhythm.
The axons of the retinal ganglion cells exit the retina via the optic nerve and cross at the optic chiasm, where the left and right visual fields of each eye are combined and redirected to opposite sides of the brain. Beyond this the optic nerve fibres continue to a few destinations in the brain such as the pretectum, which mediates processes such as pupil reflex, focus and head-related eye movements, and the superior colliculus which is involved in the control of saccadic eye movements (described in Section 3.5.1). Most fibres however continue to the two lateral geniculate nuclei (LGN), one for each side of the visual field. Each nucleus segregates information from each eye into different layers but retains a topographical map of the visual field. The geniculate cells have similar receptive fields to retinal ganglion cells that maintain the segregation of the left and right eye. Fibers from the LGN are combined in the optic radiation and project on to the primary visual cortex. It is in the primary visual cortex where, among other processes, ‘bar’ and ‘edge’ detection occurs. Bar and edge receptive fields (see Figure 3.4 and Section 3.4) are derived via selective wiring in which a cortical cell receives organised input from a set of neighbouring LGN cells. This process leads to spatial frequency and spatial phase discrimination, as described in the Sections 3.3 and 3.4.