2010 Volume 48 Issue 4 Pages 369-376
To quantify a subject's skill at visual interpolation, as in the identification of incomplete letters, we developed a new computer-based system for presenting fragmented letters of the English alphabet. Using this system, we examined the quantitative relationship between the fragmentation of letters and their correct identification in six healthy young male subjects. The 26 letters (black, 72-point, MSP Gothic) were randomly displayed one by one on the screen of a personal computer: each letter appeared once per test. Each of the letters was presented within a square (128 × 128 pixels) against a white background for 200 ms. The fragmented letters were produced by randomly removing pixels from complete letters using three different modes of elimination: removal of single dots or small and large rectangles with random rotation. The complete and fragmented letters were evaluated with respect to the density of information provided by the pixels that constituted the letters according to information theory. There was a direct correlation between the percentage of pixels removed from the complete letters and the density of information about the removed pixels. The scores for the correct identification of fragmented letters with dot elimination remained at almost 100% correct identification regardless of the percentage of pixels removed. In contrast, the correct identification scores with the elimination of large rectangles decreased with the removal of 70%, 80%, 86%, 90%, and 92% of pixels. The correct identification scores for the elimination of small rectangles fell between the scores for the other elimination modes. These results suggest that the identification of fragmented letters could be related to the recognition of structures in which imaginary lines exist between closest pixels, within some limited distance between existing pixels. The evaluation of fragmented-letter identification using this system may be useful for quantifying the capability of visual interpolation.
