Material below summarizes the article Smooth Versus Textured Surfaces: Feature-Based Category Selectivity in Human Visual Cortex, published on September 9, 2016, in eNeuro and authored by Cesar Echavarria, Shahin Nasr, and Roger Tootell.
Throughout the primate visual cortex, individual neurons and clusters of neurons respond quite strongly to specific features within a viewed image. For example, neurons within the primary visual cortex (area V1) respond strongly to lines presented at one orientation but not to other orientations. Such neurons are orientation selective. Similarly, many neurons within area MT/V5 respond strongly to stimuli moving in one direction compared to other directions. These neurons are direction selective.
The nature of such neural selectivity within a given area reflects the role that this area plays in our perception of visual stimuli. In early-stage areas, such orientation and motion selectivity was initially discovered by qualitative testing of single unit activity in response to simple visual stimuli including bars or edges. However, in many later-stage visual areas, it has been more challenging to discover the exact nature of neural selectivity using such simple stimulus comparisons, i.e. response properties are more complex.
This situation is illustrated in the lateral occipital (LO) area in the human visual cortex. LO is considered to be object selective partly because it responds more strongly to images of objects compared to images of scrambled objects. Scrambled images are typically generated by artificially cutting up each object image, then randomly positioning the cut-up portions so that the image is no longer a recognizable object. Using this criterion, area LO is considered to be object-selective and involved in object recognition.
However, this object selective interpretation of LO activity does not greatly narrow down the response properties because much of the visual world can be categorized as objects. In this study, we used a novel unbiased approach to discover which stimuli evoke the strongest response in LO. Intriguingly, this approach revealed that LO responds more strongly to stimuli with smooth surfaces compared to stimuli with textured surfaces. Moreover, this preference for smooth (compared to textured) surfaces in LO extended to the fusiform face area (FFA), which is face selective.
First we quantitatively analyzed the appearance of images depicting either intact or scrambled objects. To minimize experimenter bias, we used an image processing technique called texture synthesis that was originally developed to describe images of textures. This technique involves two stages. We measure various simple properties of a given image, which yields a set of values that describe the image. Then, an algorithm uses these values to generate a new image that has the same image properties as the original image but is not recognizable from the original image. Such images are no longer objects.
The first experiment tested whether LO activity could be modulated by the appearance of non-object stimuli. We used the texture synthesis technique to generate two sets of stimuli. The first set of stimuli consisted of images generated to match the properties of intact object images. We referred to these as texture-synthetized (TS) stimuli. The second set of stimuli consisted of images generated to match the properties of scrambled object images, which we labeled TS- stimuli. Results from a control behavioral experiment showed that naïve observers could not recognize either set of stimuli. Thus, the synthetized stimuli were matched for non-recognition.
We then conducted an fMRI experiment in which participants viewed both TS and TS- stimuli. Our hypothesis was that TS stimuli would evoke a higher level of activity within LO, compared to TS- stimuli — that LO responds selectively to certain lower-level properties in some images, compared to other images even in the absence of object recognition. Our results supported this hypothesis. Moreover, this result extended to an additional cortical area, known as FFA.
The second experiment tested whether LO activity could be modulated by the appearance of everyday objects. To test this, we used the first stage of the textures synthesis technique to generate a set of values for each of the 300 images of everyday objects. These values determined whether each object was closer in appearance to the images of intact or scrambled objects.
We then created two sets of stimuli. The first image set consisted of objects that were closer in appearance to the images of intact objects. We found that this set of objects typically had obviously smooth surfaces, and thus we referred to these as smooth objects. The second image set was complementary, consisting of objects closest in appearance to the images of scrambled objects. These objects appeared to have textured surfaces so we named them textured objects. Results from a control perceptual experiment confirmed that smooth and textured objects were easily and equally recognizable.
After, we conducted an fMRI experiment in which participants viewed smooth and textured objects. Our hypothesis was that neural activity within LO would be higher when participants viewed smooth objects compared to textured objects. Our results supported this hypothesis. Again, FFA responded analogously.
Results from both experiments showed that like LO, FFA activity is modulated by the appearance of object and non-objects. Accordingly, Experiment Three tested whether this response bias extends to images of faces, to which FFA responds selectively.
We first created two sets of computer-generated images of faces. The first set of face stimuli had a smooth complexion and appearance. The second set of face stimuli was given a textured appearance by the addition of freckles to the faces in the smooth image set. We hypothesized that neural activity within LO and FFA would be higher when participants viewed smooth faces compared to freckled faces. Results of an fMRI experiment supported this hypothesis.
Overall, our results show that LO and FFA respond selectively to specific type of non-objects, objects, and faces. Relatively smoother surfaces may facilitate object recognition in a manner not captured by our control behavioral experiments. These results refine our understanding of the nature of information processing at middle cortical stages.
Visit eNeuro to read the original article and explore other content. Read other summaries of JNeurosci and eNeuro papers in the Neuronline collection SfN Journals: Research Article Summaries.
Smooth versus Textured Surfaces: Feature-Based Category Selectivity in Human Visual Cortex. Cesar Echavarria, Shahin Nasr, Roger Tootell. eNeuro Sep 2016, 3 (5) DOI: 10.1523/ENEURO.0051-16.2016
About the Contributor
