Computational neural networks for detection of Mach bands

Visual processing models based on lateral inhibition are tested for their ability to generate Mach bands for various luminance inputs. The Huggins-Licklider model, a stimulus dependent model, is implemented on a hard-wired neural network to determine the conditions which favor Mach band production. Using this computational model, a step function luminance is shown to produce Mach bands. As the slope of the luminance get steeper, the Mach bands become more pronounced. The response comparing the negative second derivative of the luminance curve to the luminance input itself exhibits the Mach effect. This conclusion questions Ratliff´s findings (1984) that a step function luminance will not produce Mach bands