Errata in Publications

(Typos or misstatements unlikely to mislead anyone are not included here.)


In Graham & Nachmias (1971)

There is an error in Graham & Nachmias (1971),middle p. 253. In describing Fig. 2, row 3, it says that the peak height in the output of a single-channel model to a peaks-subtract compound is 1.4 times the threshold amount. However, the correct ratio is 1.54. (back to abstract in list)


In Graham (1989a, Visual Pattern Analyzers)

In Fig. 4.2 of the book Visual Pattern Analyzers ( Graham, 1989a) \ the peak height in the output of the single-analyzer model to the peaks-subtract compound is labeled as 1.4T. But the correct value is 1.54. (back to abstract in list)


In the Appendices of Graham, N. (1989b) and Sutter, Beck, Graham (1989)

In the equation given for the Gabor function near the beginning of each appendix, wx and wy are the HALF bandwidths at half-peak-height (although the accompanying text says FULL bandwidth at half). The correct equation for FULL bandwidth would have an extra factor of 4 multiplying the ln2 factor each place it appears (see table 2.2 on page 51 of Graham, 1989a).

Later in the appendix, the text correctly says that the width (or spatial extent) of the field perpendicular to the bars was chosen so that the spatial-frequency FULL-bandwidth at half-peak-height was one octave. But there is then an equation in the parentheses following this comment which is mangled. As written the equation says:

wx = wy = 2/3f

where f is the frequency of the Gabor patch (the "preferred frequency" of the filter). This is wrong. The equation was probably originally supposed to read:

wx = wy = 0.8825/((2/3)*f)

where wx and wy were supposed to be the FULL bandwidths at half height. This equation would have been better written as:

wx = wy = 1.32/f

See Table 2.2 (p. 51), Table 2.3 (p. 52) and Table 2.4 (p. 62) and accompanying text of Graham, 1989a, for fuller descriptions of the relationships between the bandwidths of the envelopes in the spatial domain and those in the frequency domain. Briefly, let wfx and wfy be the FULL bandwidths at half-peak-height of the Fourier transform of the Gabor patch, then

wx = 0.8825/wfx (Table 2.2 of Graham 1989a)

wfx = 0.67f = (2/3)*f for a one-octave bandwidth (Table 2.3 of Graham 1989a)

Combining these last two equations gives:

wx = 0.8825/((2/3)*f) = 1.32/f

(back to abstract of Graham, 1989b) (back to abstract of Sutter, Beck, and Graham, 1989)



In Graham, Sutter, Venkatesan and Humaran (1992) and Graham, Sutter, and Venkatesan (1993)

There is an inconsistency between the stated viewing distance and the spatial characteristics of the patterns in Graham, Sutter, and Venkatesan (1993) and Graham, Sutter, Humaran, and Venkatesan (1992).

We ran these experiments at two viewing distances leading to two different ranges of spatial characteristics: When the viewing distance was 0.91 meters, then 32 pixels occupied 0.67 deg of visual angle, and the fixed spatial frequency was 12 c/deg. When the viewing distance was 0.6 meters, then 32 pixels occupied 1 deg of visual angle, and the fixed spatial frequency was 8 c/deg.

Our experimental results at both viewing distances were the same (within the constraints of the usual variability) and all conclusions drawn from them would have been the same.

The fullest set of results was collected at a viewing distance of 0.91m. It is this set of results which is in the papers, and the viewing distance is correctedly identified. Unfortunately, the papers give the spatial parameters corresponding to the closer viewing distance, e.g. they say that the fixed spatial frequency was 8 c/deg where, in fact, it was 12 c/deg. However, as mentioned above, the results at 8 c/deg led to the same conclusions.

(back to abstract of Graham, Sutter, Venkatesan, and Humaran 1992) (back to abstract of Graham, Sutter, and Venkatesan, 1993)