Michael Price
ABSTRACT
The purpose of this experiment was to relate digital noise, which was an option found in most graphic programs, to the recognition of certain pictures. In this
experiment, it was expected that the digital noise, as it was decreased, would help in the
recognition of images.
Nine images were taken from America Online, a telecommunications network. These images were saved onto a floppy (3.5") disk. Using a program called Paint Shop Pro v 4. 1, these images were rendered with digital noise. The noise was added 100 percent on the clean image, saved onto the disk as another file, then, in twenty percent intervals, the noise was added and saved until it reached 180 percent, Fourteen volunteers were chosen and each were asked if they could identify the 180% noise image. If they could not, the file with 160% noise was brought up and then again each were asked if they could recognize it. This process repeated until the noise percentage was zero. If the volunteers still could not identify one of the "clean" images, they were replaced by another volunteer.
As it turns out, the results varied for each celebrity. Some volunteers needed mostly the "clean" image of a celebrity. Some did not need to go far in the decreasing of noise in order to identify them. In all, the recognition depended on the features of a certain celebrity's face.