Retinal Neuronal Coding

The retina selects, compresses and optimizes incoming information by selecting new and unpredicted information. These selection mechanisms are much more sophisticated than originally assumed and tasks previously thought to occur in the cortex are already performed in the retina. Therefore, understanding the retinal coding principles is fundamental for studying visual coding in the rest of the brain.

A considerable amount of processing needed for this, occurs in the outer retina. We showed that cones compress the incoming stimulus such that the skewed intensity distribution, present in natural stimuli, becomes a more efficient symmetrical distribution. We elucidated a novel form of contrast adaptation in cones, in spite of the established common opinion. When processing high contrast images, cones have a band pass filter characteristic. When processing contrast images, they display a low pass filter characteristic. We subsequently showed that the negative feedback pathway from horizontal cells to cones generates instantaneous color constancy (see images below). This work challenged the long standing dogma that color constancy is a feature generated in V4 of the visual cortex.

The left image is the original image. The apple in the middle is obviosly red. When putting a blue filter on top of the apple, the apple becomes realy blue (middle image). However, when we extend that same filter over the whole image, the apple appears red again (right image).

Finally, we identified, for the first time, that the signals from the cones are relayed to two general types of bipolar cells, each with its own specific intensity dependence. One bipolar cell type is sensitive over a broad intensity range and is especially suited to detect global luminance levels. The other type is an intensity opponent bipoalr cell that changes its response polarity depending on the luminance of the stimulus. They are especially sensitive to detect small fluctuations in light intensity. We will continue studying these mechanisms by using information theory, and an “ideal observer approach” to quantify the performance of the retina.