Retina - Spatial Encoding

Spatial Encoding

The retina does not simply send a picture to the brain. The retina spatially encodes (compresses) the image to fit the limited capacity of the optic nerve. Compression is necessary because there are 100 times more photoreceptor cells than ganglion cells as mentioned above. The retina does so by "decorrelating" the incoming images in a manner to be described below. These operations are carried out by the centre surround structures as implemented by the bipolar and ganglion cells.

There are two types of centre surround structures in the retina—on-centres and off-centres. On-centres have a positively weighted centre and a negatively weighted surround. Off-centres are just the opposite. Positive weighting is more commonly known as excitatory and negative weighting is more commonly known as inhibitory.

These centre surround structures are not physical in the sense that one cannot see them by staining samples of tissue and examining the retina's anatomy. The centre surround structures are logical (i.e., mathematically abstract) in the sense that they depend on the connection strengths between ganglion and bipolar cells. It is believed that the connection strengths between cells is caused by the number and types of ion channels embedded in the synapses between the ganglion and bipolar cells. See Receptive field for figures and more information on centre surround structures.

The centre surround structures are mathematically equivalent to the edge detection algorithms used by computer programmers to extract or enhance the edges in a digital photograph. Thus the retina performs operations on the image to enhance the edges of objects within its visual field. For example, in a picture of a dog, a cat and a car, it is the edges of these objects that contain the most information. In order for higher functions in the brain (or in a computer for that matter) to extract and classify objects such as a dog and a cat, the retina is the first step to separating out the various objects within the scene.

As an example, the following matrix is at the heart of the computer algorithm that implements edge detection. This matrix is the computer equivalent to the centre surround structure. In this example, each box (element) within this matrix would be connected to one photoreceptor. The photoreceptor in the centre is the current receptor being processed. The centre photoreceptor is multiplied by the +1 weight factor. The surrounding photoreceptors are the "nearest neighbors" to the centre and are multiplied by the -1/8 value. The sum of all nine of these elements is finally calculated. This summation is repeated for every photoreceptor in the image by shifting left to the end of a row and then down to the next line.

The total sum of this matrix is zero if all the inputs from the nine photoreceptors are the same value. The zero result indicates the image was uniform (non-changing) within this small patch. Negative or positive sums mean something was varying (changing) within this small patch of nine photoreceptors.

The above matrix is only an approximation to what really happens inside the retina. The differences are:

1. The above example is called "balanced". The term balanced means that the sum of the negative weights is equal to the sum of the positive weights so that they cancel out perfectly. Retinal ganglion cells are almost never perfectly balanced.
2. The table is square while the centre surround structures in the retina are circular.
3. Neurons operate on spike trains traveling down nerve cell axons. Computers operate on a single Floating point number that is essentially constant from each input pixel. (The computer pixel is basically the equivalent of a biological photoreceptor.)
4. The retina performs all these calculations in parallel while the computer operates on each pixel one at a time. There are no repeated summations and shifting as there would be in a computer.
5. Finally, the horizontal and amacrine cells play a significant role in this process but that is not represented here.

Here is an example of an input image and how edge detection would modify it.

Once the image is spatially encoded by the centre surround structures, the signal is sent out the optical nerve (via the axons of the ganglion cells) through the optic chiasm to the LGN (lateral geniculate nucleus). The exact function of the LGN is unknown at this time. The output of the LGN is then sent to the back of the brain. Specifically the output of the LGN "radiates" out to the V1 Primary visual cortex.

Simplified Signal Flow: Photoreceptors → Bipolar → Ganglion → Chiasm → LGN → V1 cortex