The Restricted Focus Viewer (RFV) is a computer based tool for tracking visual attention. Its design is in part based on the human visual system, which can only focus on objects at the centre of the visual field. The region surrounding this area of sharp focus is still perceived, but the further from the centre of the visual field an object is, the more coarse is the perception of it.
The design of the RFV attempts to reflect this idea through the use of image blurring. The RFV displays a blurred stimulus image on a computer monitor, allowing the participant to see only a small region of the stimulus in focus at any time. The region in focus, called the focus window, can be moved using the computer mouse. The RFV records what the participant is focusing on at any point in time, and the data can be played back using the Replayer program.
The focus window of the RFV is the region in which the stimulus is visible in
full detail. In order for the focus window to look `natural', a graded blurring
effect is needed, such that the transition from blurred to focus appears smooth and
seamless.
A graded blurring effect is achieved by the technique illustrated in
Figure 1.
The outer rectangle defines the stimulus area
which is fully blurred. The innermost box is the region of focus. Surrounding
this focus region are three transition regions. Each transition region is
slightly more blurred than the last, so that there is only a subtle difference
between neighbouring regions. The overall result is the appearance of a smooth
transition from the region of the image in focus, to the region which is fully
blurred. Using the mouse to move the focus window therefore moves not only the
focus region, but also the three transition regions.
Figure 1. Regions of the stimulus used to achieve the
graded blurring effect.
It is clear then that apart from the stimulus image in full focus, four other
images are needed for the different levels of image blurring in the
blurred and transition regions. At the centre of the focus window there is
also a small black dot, to allow users to keep track of the focus window
location when it is centred on an empty region of the image.
The aim of the blurred region is that it should still be possible to perceive
the broad structure of the original stimulus image. The individual
components and finer details in the stimulus should be indiscernible,
requiring the user to move the focus window over that area of the stimulus
in order to determine exactly what is there. However, it should not be so
blurred that the user has difficulty in navigating from one stimulus component
to another. Figure 2
gives an example with an algebra expression as
the stimulus, and the corresponding blurred image in which the symbols are
indiscernible.
Figure 2. Example of a visual stimulus and its corresponding
blurred image.
Different kinds of images have different spatial properties. Thus, different
techniques are required in order to produce a blurred image that obscures
the finer details, while still allowing the general form of the image to perceived.
Since a human observer is needed to verify if an appropriate level of image
blurring has been obtained, the RFV does not do any image blurring itself, but
rather reads in image files that have already been blurred.
For each stimulus to be presented, five image files need to provided. These are
the image proper (that is, the image in full focus), the image with the
appropriate level of blurring for the innermost transition region, the images
for middle and outermost transition regions, and finally the fully blurred
image. Figure 3
gives an example of the five images used to
present the stimulus shown in
Figure 2, with the fully focused
image at the top and fully blurred image at the bottom. Each of the three
transition region images in between are slightly more blurred than the image above it.
The RFV program dynamically combines these images to produce a smooth transition
from the blurred region to the region in full focus (using the method described in the
previous section). Figure 4
gives two examples of the focus window in different positions over the
stimulus shown in Figure 2.
Figure 3. An example of the five images needed to present a stimulus.
Figure 4. Two examples of the focus window on different regions of
the stimulus.
As was mentioned before, different techniques are required to successfully blur
different types of images. Consider for example, mathematical equations and
circuit diagrams as stimuli. Mathematical equations usually are composed of a
closely packed group of symbols. A standard blurring algorithm, which can be
found on most modern computer graphics programs, is generally sufficient to
successfully blur such a stimulus. (For example, on systems running
Windows, the program Paint Shop Pro has a filter which allows images to be
blurred. On Unix systems, the program XV has a blur algorithm.) Different
levels of blurring can be achieved by varying the blur radius, or by running
the blur algorithm on the image more than once.
Circuit diagrams on the other hand usually have large empty regions, and single
lines representing wires that are hard to blur. One approach to effectively
blur such images is to pixelize the image first (that is, decrease the image
resolution without changing the image size, which again can be done by using a
pixelize algorithm found on most computer graphics programs), and then run a
standard blurring algorithm. By varying the size of pixelization, and the
amount of blurring, and running these algorithms multiple times on a stimulus
image, it is possible to obtain various levels of blurring that allow the
general form of the diagram to still be perceived, but leave the individual
components indiscernible.
Another feature that was implemented so that the RFV would more accurately mimic the
way humans perceive visual stimuli is motion blur. If the user of the RFV moves
the mouse at high speed (that is, over a large distance on the screen in a
small amount of time), the focus window will not achieve full focus. Once the
user reduces the speed of the mouse motion back to below a certain threshold,
or stops moving the mouse completely, full focus in the focus window will
return. This feature helps in defining the temporal boundary between fixations
and movements.
When the focus window is stationary or moving slowly, all of the regions listed
in Figure 1
are present. During motion blur however, only
the outermost transition region is added to the blurred stimulus. Because this
region has less blurring than the rest of the image, the user is still able to
track the location of the focus window on the stimulus. However, it is not
possible to determine the finer details of that location without slowing or
stopping the mouse. Only then will full focus be available.
When designing an experiment that will use the RFV, there are several important
parameters that the experimenter must set. Below are some basic heuristic
guidelines for setting these parameters. These are suggestions based on previous
experience in using the RFV, but by no means are they intended to be strict
rules for parameter settings.
3.1 The Focus Window
3.2 Stimulus Images and Blurring
3.3 Motion Blur
3.4 Guidelines for Setting Parameters
Level of Blurring
The goal of blurring the stimulus is to generate an image where it is difficult
to identify the finer details. The minimum level of image blurring should be
sufficient that any two stimulus elements are indistinguishable, and that the
connections between elements cannot be established. It should only be
possible to accurately identify an element if that element is in the focus
region. Of course, for different types of stimuli the size of the elements will vary,
and thus different levels of blurring will be necessary. The maximum level of
blurring should still allow identification of the stimulus boundaries (at least
the convex hull). This is to ensure that participants can still navigate from
one region of interest in the stimulus to another.
Focus Window Size
The central focus region of the focus window should allow identification of a
single element of the stimulus. Thus it should not be so small that identification
is difficult when the focus window is centred over an element. It also should not
be so large that it allows the simultaneous identification of two or more
neighbouring elements, since this will make it difficult for the experimenter
to determine which element the user was focusing on. Generally, the focus region
should be slightly smaller than the bounding box of a typical stimulus element.
Each transition region should be only slightly larger than the region within it.
The role of the transition region is to provide a smooth transition from focused to
blurred, and to indicate the direction of neighbouring connected elements.
Motion Blur Speed
The motion blur feature allows for a distinction to be made between movements
and fixations. The threshold speed
should not be so high that users can use a `brass rubbing'
strategy. That is, identifying the stimulus by rapidly moving the focus window
over it. However, it should also allow for slow navigation between connected
stimulus elements. The mouse speed at which motion blur onset occurs will depend
largely on the type of stimulus and the nature of the task. For example, if the
stimulus is a circuit diagram, than the motion blur speed should be high to
allow the user to trace the path of the wires without losing focus. If instead
the stimulus is mathematical equations, then motion blur onset should occur for
much slower mouse speeds.
Back to Contents |
Previous Section |
Next Section