Summary: IV. Curve detection: semi­global
methods
ffl Instantiation parameters ` directly in terms of constraints the locations of the
n points along the curve.
`Z = ` = (` 1 ; : : : ; ` n ).
ffl Prior formulated directly in terms of the locations ­ favoring small curvature.
ffl Assume data between pairs of consecutive points on curve independent.
Model data along curve and estimate relevant parameters.
ffl Initialization through two points on the curve.
ffl Global optimization using dynamic programming.
IV.1 The Bayesian model
ffl Transform data into collection of point processes.
ffl Curve can either traverse a pixel x at one of A different angles: ff(x) = a; a =
1; : : : ; A
or no curve traverses the pixel: ff(x) = 0.
ffl X a (x); a = 1; : : : ; A collection of local binary features (see below).
X a (x) is more likely to be 1 if ff(x) close to a.
32

IV. 1­D GLOBAL METHODS 33

Source: Amit, Yali - Departments of Computer Science & Statistics, University of Chicago

Collections: Computer Technologies and Information Sciences