Figure 11

Flexibility of the nonlinearities of the texture sensors as a function of the semisaturation constants. Here flexibility is modified in different rows by changing the semisaturation b in the divisive normalizations (\(100 \times b\) in the first row and \(0.01 \times b\) in the third row). In these plots, we show the responses of texture sensors tuned to low, medium, and high frequencies (left, center, and right columns respectively) for multiple images (each point in the scatter plots is one image). The central row shows the response of the sensors in the baseline model. Note how the nonlinearity is image-dependent, and hence there is no a single response curve, but rather a scatter plot. As the flexibility is increased (from the top row to the bottom row), the response goes from a fixed curve to a wider family of curves