Second-order Pattern Process

Second-Order Pattern Mechanisms

It is well documented that the visual system is sensitive to both the first-order and the second-order variations in an image. The first-order information usually refers to the spatiotemporal modulation in luminance or chromaticity and it can be specified as luminance changes across space. Such modulation has a corresponding energy in the Fourier domain and is detectable by a linear operator whose receptive field contains excitatory and inhibitory regions similar to that of neurons in the primary visual cortex. On the other hand, the second-order information can be described as the variation in the changes of luminance with the same mean luminance across space. The second-order information has no corresponding energy in Fourier domain and is not detectable by a simple linear receptive field. In the literature, linear-nonlinear-linear (LNL) models are commonly used to explain the visual behavior to a second-order stimulus. This model involves a band of linear filters that is responsive to the elements of the image. The responses of the linear filters are then undergoing a nonlinear transform before sending to the 2nd-stage linear filters.

My project focused on the underlying mechanism for the second-order pattern stimuli by using pattern masking paradigm and adopted Foley's (1994) divisive inhibition model to investigate the nonlinear process of pattern masking results for second-order stimuli.

• • Huang, P. C., & Chen, C.-C. (2009). Pattern masking investigations of the second-order visual mechanisms. Proc. SPIE, 7240, 724016. [link]

  • Huang, P. C., & Chen, C.C. (2014). A comparison of pedestal effects in first- and second-order patterns. [link] Journal of Vision, Jan 10;14(1). [full text]

Usually, texture is also viewed as a second-order information due to the fact that the mean luminnance across the adjacent area is the same but the variance of its feature is different. My master's project focused on the texture-defined edge. Boundary is defined as the area that surface changes quickly. To form an edge representation, two processes are involved: 1) count the number of features that the filters responded. 2) calculate the feature gradient from the first step input, and define the peak as an edge.The project was to investigate the second stage mechanism of edge defined by texture.