Abstract. A prominent model of visual motion detection is the so-called correlation or Reichardt detector. Whereas this model can account for many properties of motion vision, from humans to insects (review, Borst and Egelhaaf 1989), it has been commonly assumed that this scheme of motion detection is not well suited to the measurement of image velocity. This is because the commonly used version of the model, which incorpo- rates two unidirectional motion detectors with opposite preferred directions, produces a response which varies not only with the velocity of the image, but also with its spatial structure and contrast. On the other hand, information on image velocity can be crucial in various contexts, and a number of recent behavioural experi- ments suggest that insects do extract velocity for navigational purposes (review, Srinivasan et al. 1996). Here we show that other versions of the correlation model, which consists of a single unidirectional motion detector or incorporates two oppositely directed detec- tors with unequal sensitivities, produce responses which vary with image speed and display tuning curves that are substantially independent of the spatial structure of the image. This surprising feature suggests simple strategies of reducing ambiguities in the estimation of speed by using components of neural hardware that are already known to exist in the visual system. 1 Introduction The simplest way in which a visual system can determine how fast and in what direction an object travels would be to determine how long the object needs to cover the distance between two given points. Indeed, `feature- tracking' mechanisms have been proposed along these lines to explain motion vision (Braddick 1980; Ullman 1983) and are sometimes used in machine vision (Murray and Buxton 1990). The disadvantage of such schemes, however, is that they need to identify the object, or features within it, before carrying out the tracking. In another class of models, this problem is avoided by using information on local spatiotemporal changes of intensity to measure velocity. In one subclass of these `intensity-based' models, the so-called gradient models, image speed is determined by computing the ratio of the local temporal and spatial gradients of intensity (Fennema and Thompson 1975; Limb and Murphy 1975). Gradient models have the property that they measure the speed of an image independently of its spatial structure. Modi®cations of this scheme have been proposed for measuring image velocity in two dimen- sions and for overcoming problems associated with low and sparsely distributed contrasts (Hildreth and Koch 1987; Johnston et al. 1992; Srinivasan 1990). Another subclass of the intensity-based models is the so-called Reichardt or correlation detector, which ex- tracts a motion signal from the spatiotemporal correla- tions that are present in the moving image (Reichardt 1957). In this model, which has been very successful in describing motion sensitivity in animal vision from in- sects to primates (review, Borst and Egelhaaf 1989), the signal from one input unit A is delayed or temporally low-pass ®ltered and multiplied with that from a neighbouring input B (see Fig. 1). As a consequence of this structure, the model produces a strong output only when the image moves in the direction (A ! B). The standard design of the correlation model, the so-called balanced version, subtracts the output of this network from that of an anti-symmetric one which multiplies the signal from A with the ®ltered signal from B. The re- sulting scheme produces a positive response when the image moves in the direction (A ! B), and a negative response when the image moves in the direction (B ! A). Setting these two anti-symmetrical `half-detectors' in opposition has the virtue of cancelling out, by subtrac- tion, the direction-unspeci®c components of the re- sponse and makes the output of the overall network highly selective for the direction of motion (Borst and Egelhaaf 1990). Direction-unspeci®c response compo- Biol. Cybern. 80, 109±116 (1999) Speed tuning in elementary motion detectors of the correlation type J.M. Zanker 1 , M.V. Srinivasan 1 , M. Egelhaaf 2 1 Centre for Visual Sciences, RSBS, Australian National University, GPO Box 475, Canberra, ACT 2601, Australia 2 Lehrstuhl fuÈr Neurobiologie, UniversitaÈt Bielefeld, Postfach 100131, D-33501 Bielefeld, Germany Received: 30 April 1998 / Accepted in revised form: 18 September 1998 Correspondence to: J.M. Zanker