Spatial frequency processing is what is thought to be one of the earliest stages of visual processing. Neural mechanisms called channels or filters specialise in the simplest components of the visual scene: sinusoidal waves. In doing this, early visual processing mechanisms for edge detection, follow the logic of the Fourier theorem in physics. According to this, any complex pattern is the sum of simple sinusoidal components of varying frequency, amplitude, orientation and phase. Each spatial frequency channel in early vision detects a range of sinewave component frequencies in the visual scene and sends this information to higher levels of analysis to be combined with signals from other channes to form edges. This in turn leads to object recognition, identification, and all the other functions that the visual system has to perform.

This figure shows two simple visual patterns: a sinewave pattern (upper) and a squarewave pattern (lower). Note how in the sinewave pattern the light and dark bars change smoothly (continuously) as plotted on the right. Conversely the squarewave pattern consists of light and dark bars changing suddenly.

In this figure, the three patterns of bars changing from light to dark sinusoidally, are stimuli used in visual psychophysics experiments, called sinewave gratings. The top grating follows a sinewave pattern with high amplitude (high contrast - large difference between light and dark). This middle grating is a low amplitude (contrast) one. The bottom grating has a lower frequency (thicker bars) than the first two.

This figure demonstrates the way in which a complex pattern is formed by summing simple sinewave components as the Fourier theorem suggests. The component sinewaves increase in frequency and decrease in amplitude by a certain proportion. In fact, these are called harmonic frequencies.

When an infinite number of harmonics are combined, a squarewave is formed. So, ironically, one of the most complicated patterns in vision is a squarewave!

Interestingly, the most complicated pattern you can find is a dot!

This is another demonstration of complex patterns being the sum of simple sinusoid components. In this case the two complex patterns are called compound gratings and they are formed by summing gratings of differing frequencies..

Here, a "plaid" is formed by combining gratings of differing orientations.

And now see how a complex pattern (in this case a picture of Einstein) consists of different spatial frequencies. The top-left image shows lowest frequencies in the picture, the top-right a higher range of frequencies, and the bottom-right the highest.

My research in this area started during my post-doc position, working with Paul Sowden at Surrey University and Philippe Schyns at Glasgow University in Britain. Our research was a psychophysical follow-up to earlier evidence reported by Philippe Schyns and colleagues, suggesting that spatial frequency channels may function in a flexible fashion, dictated by task demands. Such evidence and others suggested that spatial frequency processing may be influenced by cognitive functions such as sensitisation, categorisation and expectations.

These are stimuli used in a study by Philippe Schyns, one of my collaborators, and Aude Oliva. These are the kinds of stimuli we use in some of our studies.

Try it yourself! Look at the images above close up and then from a distance of a few meters away from the screen. You will see that the faces change from a man to a woman and that the facial expressions change from neutral to smiley or angry to neutral. This is because there are actually two faces in each of these pictures: one consisting of high spatial frequencies (like the bottom-right picture of Einstein above) and one of low frequencies. At short distance, you are able to see high frequencies (fine detail), while at long distances, the fine detail face disappears and you now see the low spatial frequency (coarse) face in the image.

We conduct experiments exploring the issue of whether top-down attentional modulation of spatial frequency channels may be responsible for the effects reported by Schyns and others. In other words, we have investigated the locus of such effects, testing for a range of indications of processing as low-level (as early) as spatial frequency channels, such as tuning bandwidth and retinotopic specificity.

To read more on the subject, please follow the link to publications on the left.