Role of low and high spatial frequencies in the face-selective responses of neurons in the cortex in the superior temporal sulcus in the monkey.

There are neurons in the cortex in the anterior part of the superior temporal sulcus of the macaque monkey with visual responses selective for faces. One aim of the present study was to analyze further the information which leads them to respond, by measuring their responses to parametrically filtered stimuli. The responses of 32 such single neurons were measured to faces which were digitized, lowpass filtered at spatial frequencies of 2, 4, 8,...128 cycles/face, highpass filtered at frequencies of 4, 8,...64 cycles/face, and presented in random sequence using a video framestore. It was found that many of the neurons could respond to blurred images of faces, with a mean frequency at half-maximum amplitude of the neuronal response to the series of lowpass filtered images of faces of 3.3 cycles/face. Almost all the neurons had lowpass cutoff frequencies defined in this way below 8 cycles/face. Many of the neurons could also respond to images of faces in which the only information remaining was a limited amount of high spatial frequency edge information. The mean frequency at half-maximum amplitude of the neuronal response to the series of highpass filtered images of faces was 29.7 cycles/face. Almost all the neurons had highpass cutoff frequencies above 8 cycles/face. Thus, many of the neurons could respond to a lowpass and a highpass filtered image of a face even when these had no spatial frequencies in common. The mean separation between the lowpass and highpass cutoff frequencies was 3.2 octaves. For comparison, face recognition in man can be performed with images which contain only information up to 8 cycles/face, or with highpass filtered images which contain only information down to 8 cycles/face. The response of the neurons was not always a smooth function of frequency, but could decrease as higher frequencies were included in the lowpass filtered images of faces, or as low frequencies were included in the highpass filtered images of faces. This indicates that information in certain frequency bands was able to inhibit these neurons. This was particularly likely to occur for the non-optimal face stimulus for a given neuron, indicating that the selectivity of these neurons to different faces was a combination of the excitation produced by some information in faces and inhibition produced by other.