Typically Young's fringe pattern automatic analysis from a double-exposure image (e.g. a photograph) passes through an indirect processing stage on some intermediate parameter domain. Here, we propose a method based on a complicated image processing technique, operating directly with the source fringe image pixels, and providing remarkable accuracy and computational time. This method is intended for laser speckle velocimetry (LSV), particle image velovimetry (PIV), and digital image velocimetry (DIV) applications. Assuming a common fringe pattern model, we 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 quantitatively attributed with the corresponding eigenvectors. The algorithm has been tested on real patterns as well as on a set of artifically simulated images with pre-defined characteristics.