Premium
Efficiency of human pattern recognition as a function of the range of size randomization
Author(s) -
Näsänen Risto,
Syväjärvi Antti,
Rovamo Jyrki
Publication year - 1997
Publication title -
ophthalmic and physiological optics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.147
H-Index - 66
eISSN - 1475-1313
pISSN - 0275-5408
DOI - 10.1046/j.1475-1313.1997.9780748x.x
Subject(s) - ranging , numeral system , artificial intelligence , contrast (vision) , computer science , pattern recognition (psychology) , mathematics , range (aeronautics) , computer vision , statistics , engineering , telecommunications , aerospace engineering
We measured pattern recognition efficiency for numerals in spatial noise. The numerals of Century Gothic typeface were filtered with a two octaves band‐pass filter. Different magnifications (ranging from one to four) of the filtered numerals were used. The contrast energy across magnifications was constant. In the experiment, the different magnifications were selected for presentation in a random order. Using an n ‐alternative forced‐choice method, recognition thresholds were measured as a function of the range of magnifications. The results obtained from human observers were compared with those of an ideal (‘template matching’) pattern recognition observer. The thresholds of the ideal observer were obtained by computer simulations. Human results were expressed in recognition efficiency defined as the ratio of the ideal and human energy thresholds. With no size uncertainty the efficiency was about 20%, and when the possible magnifications varied by a factor of four, the efficiency was about 75%. Thus, human recognition efficiency became better with increasing size uncertainty. The result may be interpreted so that human vision automatically assumes that the retinal sizes of objects are not fixed but vary, for example, because of varying viewing distance.