In figure segregation that may accompany figural completion, only 1 in
terpretation is dominant for some figures, whereas several different i
nterpretations can be selected almost equally for other figures. This
article addresses 2 problems in explaining this fact: (a) how to defin
e quantitative measures for features (including gestalt principles of
grouping) and (b) how to estimate the probability of each interpretati
on on the basis of these definitions. Quantitative definitions are giv
en to 7 features involved in figure segregation and completion: the re
lative number of corners, good continuation, symmetry, curvature const
ancy, convexity, coincidence, and similarity. Each feature's value can
be calculated from the total curvature function of the contour. The p
robability of each interpretation drawn by the participants is calcula
ted by applying linear multiple regression analysis. Experimental resu
lts indicate that the proposed method can estimate human performance o
n a figure segregation and figural completion task fairly well.