Mechanisms for sensitivity
We have seen how signal detection theory proposes that the presentation of a stimulus, such as a random-dot kinematogram, is assumed to evoke an 'internal response' that, when combined with the observer's criterion, forms the basis of a perceptual decision. The mechanisms that produce the 'internal response' are not required to be specified — that is, we do not need to know how the physical stimulus is transformed into the 'internal response' in order to use signal detection theory. However, an understanding of such mechanisms can be informative of how the visual system goes about processing the physical stimulation.
In this section, we will consider a potential way in which the visual system might process visual stimulation to permit sensitivity to motion.
Requirements
First, we need to think about what sort of properties we would like our motion detection system to possess. We can gain some insight by again looking at the space-time representation, such as of the 100% coherence rightwards global motion stimulus depicted in the figure below.
We can then compare this with the space-time representation of a 'noise' (0% coherence) stimulus shown in the figure below:
The main difference between the two is that the 'signal' space-time representation contains many instances of patterns that look like ╲ . These 'streaks' are caused by the coherent rightwards motion in the stimulus. In contrast, the 'noise' space-time representation contains lots of blobs and lines pointing in different directions.
Hence, it is the presence of the 'streaks' that primarily distinguishes the 'signal' and 'noise' space-time representations. We would therefore want the response of our motion detector to be sensitive to the 'streaks' evident in the space-time representation of the 100% coherence stimulus.
Put back into the language of the physical stimulus, that means that we would want the response of our motion detector to depend on:
- The motion direction.
- For our example, it should give a large response to rightwards motion and smaller responses to other directions.
We will also impose two additional requirements that any good motion detector should satisfy:
- The motion speed.
- It should give a large response to the speed of our stimulus (indicated by the slope of the 'streaks') and a smaller response to slower or faster speeds.
- Whether the stimulus is actually moving.
- It should not give a large response to a stationary input.
A simplified scenario
To simplify the investigation of our motion detection circuit, we will use a moving bar of light rather than a random-dot kinematogram.
With this simplified stimulus, we would expect a strong response to the space-time representation shown in the figure below:
We would not expect a strong response to each of the following:
What does the above space-time representation show?
_
What does the above space-time representation show?
_
What does the above space-time representation show?
_
What does the above space-time representation show?
_
Motion detection circuit components
We will now think about what components we will need to use to implement our motion detection circuit. We will call these 'neurons', in acknowledgement that the motion detection system that we are considering would actually be implemented by networks of neurons in the brain. However, this labelling is rather loose — we will not be considering any of the biophysical properties of actual neurons, or even any of the principles of neural responses that you will learn about in this course.
For our motion detection circuit, we will use the following types of 'neurons':
- Intensity neuron
- This neuron signals the intensity of the visual input in a particular location in space (its receptive field).
- Delay neuron
- This neuron receives input from an Intensity neuron and stores it for a short period of time before retransmitting. It also adapts, and will only retransmit its input if it has not recently transmitted.
- Comparator neuron
- This neuron receives multiple inputs and performs a multiplicative operation. That is, if any of the inputs are weak (~0) then the neuron does not output a signal — all the inputs are required to be active (~1).
- Summation neuron
- This neuron receives input from other neurons and sums them together.
Circuit structure
To implement our rightwards motion detector, we organise the components in the way shown in below. This creates a form of 'delay-and-compare' circuit.
How does this work?
- Starting at the top, we have two Intensity neurons. One of the neurons has its receptive field on the left side of space and the other has its receptive field on the right side of space.
- The Intensity neuron on the left connects to a Delay neuron.
- The output of the Delay neuron and the right Intensity neuron then connect to a Comparator neuron.
- The output of the Comparator neuron is then passed to the Summation neuron — the output of this Summation neuron is the output of the motion detector.
Circuit operation
Why does this circuit structure meet our requirements? Let's talk through how it would process the space-time representation of rightwards motion that we introduced earlier.
First, here is the motion that we are seeking to detect:
Now, let's depict the spatial locations that our Intensity neurons will be sensitive to. As shown below, the left Intensity neuron responds to intensities in a small region to the left of centre and the right Intensity neuron responds to intensities in a small region to the right of centre.
The left Intensity neuron will thus be responsive at around 0.75 seconds into the video and the right Intensity neuron will be responsive at around 1.25 seconds into the video. To put it another way, the right Intensity neuron responds 0.5 seconds after the left Intensity neuron.
This different response timing is critical for our motion detector, because the Comparator neuron requires both inputs to be active at the same time. That is where the Delay neuron comes in — if we ask it to 'hold on to' the signals coming from the left Intensity neuron for 0.5 seconds, then the active input to the Comparator neuron will arrive simultaneously and produce a strong response in the Summation neuron.
To summarise the operation of the circuit:
- At ~0.75 seconds into the video, the left Intensity neuron becomes active.
- This activity from the left Intensity neuron is passed to the Delay neuron, which stores the activity for 0.5 seconds.
- At ~1.25 seconds into the video, the Delay neuron then passes this stored activity on to the Comparator neuron.
- At the same time (~1.25 seconds into the video), the right Intensity neuron becomes active and passes this activity directly to the Comparator neuron.
- Hence, the Comparator neuron is receiving activity from both of its inputs at around the same time (~1.25 seconds after the video started). Because both inputs are active, the Comparator neuron passes this activity on to the Summation neuron
- Because the Summation neuron forms the output of the motion detector, we would hence obtain strong levels of 'internal response' from this motion stimulus.
Does this circuit meet our other requirements? We will discuss further in the "low-level motion" lecture.
Summary
In this section, we wanted to gain some insight into the mechanisms that might underpin the 'internal response' in signal detection theory. By defining a few key 'neural' components, which affect the processing of intensity signals in different ways, we can organise a circuit that satisfies our requirements for a rightwards motion detector at the speed of our stimulus. This provides a potential mechanism by which the biological visual system might go about producing an 'internal response' to a stimulus.