Typically Young's fringe pattern automatic analysis from a double-expo
sure image (e.g. a photograph) passes through an in&rect processing st
age on some intermediate parameter domain. Here, we propose a method b
ased on a complicated image processing technique, operating directly w
ith the source fringe image pixels, and providing remarkable accuracy
and computational time. This method is intended for laser speckle velo
cimetry (LSV), particle image velocimetry (PIV), and digital image vel
ocimetry (DIV) applications. Assuming a common fringe pattern model, w
e introduce a pre-processing stage to improve significantly the fringe
discernment. A dynamic thresholding segmentation scheme, adjusted to
the fringe spatial structure, follows to localize the fringes being qu
antitatively attributed with the corresponding eigenvectors. The algor
ithm has been tested on real patterns as well as on a set of artifical
ly simulated images with pre-defined characteristics.