[accessibleimage] Re: Haptic rendering of images

Hi Peter,

Yes, this sometimes works quite well, and that is exactly where I used the gradient-based forces for in case > of the Logitech Wingman Force Feedback mouse. However, typical images contain lots and lots of tiny details and associated "valleys" for those snap constraint > forces, such that one gets confused about unimportant edges versus important contours. It is probably solvable > with much more advanced contour detection algorithms > that discard the noise of many local
edges, but I did not further investigate this because it is > far from trivial to design a good robust edge-prioritizing > algorithm.

Agreed. One useful starting point may be to run some psychological studies aimed at trying to determine the key characteristics that people use to identify objects that form part of images and how they differentiate these objects from the rest of the sensory stimuli. I suspect that differential thresholds are used in most cases, but can think of cases where differential thresholds alone would not be sufficient.

An easy starting point to validate or invalidate this idea might be to just use The vOICe as-is with a Logitech Wingman Force Feedback mouse, and import image files that have been manually preprocessed for leaving in > only relevant edges - probably much like was done with > the Virtouch mouse educational material. It
would give some indication of whether this approach would yield results that are sufficiently intuitive to warrant the effort of designing an appropriate and automatic "cartooning" algorithm.

Thanks for the suggestion, it's something I may look into further in the future if I get time to investigate life like images.

Will


Other related posts: