The Feature Extraction of Nonparametric Curves Based on Niche Genetic Algorithms and Multi-Population Competition
W Wei,Q Wang,H Wang,HG Zhang
DOI: https://doi.org/10.1016/j.patrec.2004.10.027
IF: 4.757
2005-01-01
Pattern Recognition Letters
Abstract:This research presents an approach utilizing niche genetic algorithms (NGA) other than Hough transform (HT) in detecting nonparametric curves or undefined shapes in a binary image. The optimum curve can be concluded from the evolutions of two populations, which are separately coded along columns and rows, or from multi-population competition. In order to extract the most probable curve as human visualization does, the fitness function based on the human visual tradition model is introduced for the fitness evaluation. The NGA-based curve feature extraction approach has many unique characteristics compared with the HT method, such as the ability to obtain the trajectory and length of nonparametric curves, high convergence speed, and implicit parallelism. For NGA-based curve extraction, this paper offers detailed analysis in the construction of fitness function, NGA, multi-population competition, population reservation, and comparison with Hough transform.