The colored line segments are linearity values calculated by quadrupole convolution.
Quadrupole convolution has no adjustable parameters. No if statement is required.
The quadrupole convolution clearly shows that the bumps have linearity perpendicular to the nearly-vertical stroke. The perpendicularity is expressed by the fact that the linearity passes through zero at the bases of the bumps. Therefore, by objective standards, this is indeed the letter “t”.
When we fiddle with this image, we find that the indication of horizontality goes away just about at the time that the cross-bar starts looking like a pair of bumps. This is not a controlled experiment, but such an experiment would help establish whether quadrupole convolution indeed occurs in the human brain.
The exit stroke at the bottom, patterned after italic fonts, causes very little horizontal linearity. Maybe that’s why humans don’t find italics hard to read.
The quadrupoles evaluated outside the letter are pretty much the same as if the letter was just a dot. This reflects the fact that the quadrupole convolution of a linear feature, measured outside the feature, is small. If one evaluated the quadrupole convolution of a line of print, the result would be the horizontality of the line of print, regardless of what characters were in it.