The fixation map is an m x n matrix with zeros where there are no fixations, and d (duration) where there are fixations. I use imgaussfilt (MATLAB) with a sigma of 37 pixels (1 degree of freedom) to generate a heat map. How do I interpret the new heat map values in terms of fixation duration in milliseconds?
Dear Pallavi, I see two aspects to your question. Fixations where and how long.
(1) Am ideal fixation is a point. Due to sampling and limited spatial accuracy, it is usually captured and stored as the number of entries in a spatial array. That's the matrix you mention. Here you have to trade accuracy (more matrix elements for increased spatial resolution), the amount of data you have (at some point you have only 1 or 0 entries in a matrix element) and the spatial accuracy of your measurement (that's typically 0.5° or more). Therefore it makes a lot of sense to lowpass filter your experimental data with a Gaussian kernel of about the width of your spatial accuracy. Then you do not pretend higher accuracy than you have. The resulting function (heat map) gives the sample distribution of fixation points.
http://jov.arvojournals.org/article.aspx?articleid=2121229 is a rigorous treatment of these ideas.
That has nothing to do with fixation duration yet. In case you did not identify individual fixations but just sampled from the eye tracker at a regular rate (that's in most cases not the desired way to go) you have fixation frequency*fixation duration as your independent variable. It sort of aggregates viewing time in a specific spatial region.
You can normalize the second map by the first to have a measure of average fixation duration. But here numerics can be tricky and you should know precisely what you do. Better is to directly measure fixation durations and average these for each spatial location.
Hope that helps, Peter
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation