Purchase this article with an account.
Károly Köteles, Rufin Vogels, Guy A. Orban; Coding of material properties in macaque inferior-temporal cortex. Journal of Vision 2004;4(8):124. doi: https://doi.org/10.1167/4.8.124.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Material properties, i.e. 3D texture, can be important to recognize objects in everyday life. It is known that macaque inferior-temporal (IT) cortex contributes to object recognition. Thus we studied (1) how IT neurons code for differences in materials (2) whether the response to images of the type of material is affected by direction of illumination and (3) how the coding of shape and 3D texture interact. The stimulus set consisted of images taken from the Columbia-Utrecht Reflectance and Texture Database (CURET) that consists of pictures of natural materials illuminated from different directions. The textures were converted to grayscale and rendered on 5 different flat shapes (3–4 deg visual angle). To test whether the 3D texture responses merely reflect the Fourier power spectrum, or, that rather the phase information is important, the same neurons were also tested with images having the same power spectrum as the originals but the phase of Gaussian noise patterns. The stimuli were presented foveally during single unit recordings from anterior IT in 2 awake, fixating rhesus monkeys. Our results indicate that texture and shape are coded independently in the majority of macaque IT neurons. The response of most IT neurons depended on the direction of illumination, so there seems to be no illumination-invariant coding of materials, neither at the single cell, nor at the population level. Fourier phase randomization strongly altered the response profile, which indicates that positional information is important in the processing of 3D textures. In an attempt to determine the underlying dimensions of the representation of materials by macaque IT, we conducted exploratory analysis — MDS, PCA, cluster analysis — on the responses to these images of materials. The responses of the population of cells to 65 textures fitted well in a low (three/ four) dimensional space. We correlated the dimensions of this representation with both filtering/energy- and texton-based models.
This PDF is available to Subscribers Only