In their classic works, Hubel and Wiesel studied the responses of Primary Visual Cortex (V1) neurons by showing bright or dark bars of light as visual stimuli. Movshon et al. used smoothed alternating bright and dark bars, sine wave gratings, as stimuli. With all such stimuli, the strength of the neuronal responses are sensitive to luminance changes across space, such in sharp edges of bars or the smooth gratings. The current concept of this suggest that there is a process of contrast normalization, that is, the effect of luminance is removed by being sensitive only to relative differences in luminance across space, while subtracting or normalizing the effect of absolute luminance (brightness). While recording from parafoveal V1 using an array of chronic electrodes (Utah array), we have found neurons whose firing rate depends on the mean luminance even after adaptation. We present sine-wave gratings at 4 orientations, and two phases, at a single average luminance for several hundred trials, followed by another set of similar gratings but with different average luminance for the next several hundred trials, and so on. One of the stimuli is blank, that is, no stimulus comes on. We present all of these stimuli at four mean luminances, from dim to bright. There seem to be two populations of neurons, those with high background firing rates (17 Hz or greater) and those with low background rates (below 8 Hz). The high background neurons might be from layer 4 of the visual cortex. Neurons in both groups show firing rates dependent on the average luminance, both for the visible gratings and for the blank stimulus. About half of the observed neurons show a significant dependence of firing rate on average luminance, that is, the neuronal firing is related to the blank stimulus. Thus, the ongoing firing rate seems to be dependent on the absolute level of brightness. Out of these mean-luminance responsive units, making up about half of our sample of neurons (80 in our sample) approximately 60 per cent show an increase in firing rate with increasing luminance, another third show a decrease while for the remaining neurons the effect is idiosyncratic. Thus, it appears that part of the neural code in primary visual cortex contains information about the absolute brightness of the scene. It is widely accepted that the other end of visual system in the rostral lateral temporal lobe, area TE, is important for perceiving complex objects, e.g., hands or faces. Bilateral removal of area TE in monkeys causes impairments in assigning images to categories. The impairments, while significant, are mild enough so that the monkeys are able to correctly assign exemplars correctly about 80% of the time, even stimuli that are mixtures from the categories, so-called morphed images. Over this year we have explored whether monkeys with bilateral area TE removals are impaired in tests of figure-ground discrimination and pattern completion. Three control monkeys and three monkeys with bilateral area TE removals were tested on a cat vs. dog categorization task comprising 440 unique stimuli per session. Monkeys were trained to touch a bar to initiate a trial, and release the bar during one of two intervals; early (signaled by a red central dot) if they identified the stimulus as cat, or late (signaled by a green central dot) if the stimulus was identified as dog. A correct response resulted in the delivery of a reward, an incorrect response led to a time-out. Once the animals performed this simple categorization task we added visual noise to the images. The monkeys were presented with interleaved trials of foreground (covering the stimulus) or background noise (11 levels, from 0 to 100%). The noise was generated by replacing randomly chosen stimulus pixels with a black or white pixel, either on the background image or on the stimulus image; this makes the image look like it has snow on a television. In the absence of noise, the performance of monkeys with TE removals was, as seen before, modestly degraded relative to that of controls. The presence of background noise did not further degrade the ability of the monkeys with TE removals to categorize stimuli, suggesting that TE is not necessary for figure-ground discrimination as configured in this experiment. However, increasing levels of foreground noise caused decreasing accuracy of categorization in both experimental and control groups. As foreground noise, that is, noise on the image itself, increased from 0 to 40% noise, the performance of control and monkeys with TE removals degraded in parallel. Beyond 60% foreground noise, control monkey performance dropped abruptly becoming indistinguishable from that of monkeys with ablations; at 90% foreground noise both groups performed at chance levels. From this we infer that pattern completion depends on two mechanisms. The first is a TE-independent mechanism that supports a basic ability to perform categorization at modest levels. The second is a TE-dependent mechanism confers a substantial advantage at lower noise levels. This makes it seem that TE is critical for recognizing objects that are partially obscured; imagine having to recognize a tiger in the foliage of a forest.
Showing the most recent 10 out of 14 publications
