Try identifying the letter in the image on the left (answer: it's an e!). How do humans process the visual input to identify letters? Is letter identification as simple grating detection mediated/constrained by frequency and orientation selective channels?
Although many computational models have been proposed to explain how human observers detect and discriminate simple signals, there are fewer computational models to explain how observers identify letters. Our work in this area has followed the work in letter identification by Pelli and colleagues (Solomon and Pelli, Nature, 1994). We have tested 13 different models (including non-channellized and channellized models) in their ability to perform as good as human observers in letter identification tasks and also consistent with the pattern of results obtained by Solomon and Pelli. We have been able to discard quite a few number of models including the all linear cross-correlators (see figure; for more details send an email for a copy of Abbey and Eckstein, 1999). We are also interested in trying to understand why letters flanking a target letter in the visual periphery degrades observer's ability (also brought to our attention by work of by Pelli and Palomares).
  Template for best-linear cross-correlator model (left image: spatial domain; right image: fourier domain). This model performs at an efficiency of 1.3 % while human observers perform at 10-13 % efficiency. Therefore this model can be rejected as a model fo human letter identification (Abbey et al., 1999)
C.K. Abbey and M.P. Eckstein, "A closer look at the visual filter mediating
letter identification," Invest Ophthalmol Vis Sci 40(4):S35, 1999. (and
hopefully written up some time soon!)
C.K. Abbey and M.P. Eckstein, "The effect of flankers on letter identification
in the periphery examined with classification images, presented at the
Opt. Soc. Am. Annual Meeting, Providence, RI, (2000)
|
