Genetically fine-tuning the Hough transform feature space, for the detection of circular objects

摘要

Despite certain inherent advantages of the Hough transform (HT), it may result in inaccurate estimates of the detected parameters, in the case of excessively noisy images. In this work, we present an original method for fine-tuning the feature space for the HT using genetic algorithms (GAs). The aim is to find a subset of features that best describe the instances of the sought shape, so that the HT accumulator is contaminated the least by noisy information. A hybrid GA/HT system is configured, by embedding the HT module into the GA, which simultaneously performs feature space fine-tuning and shape detection. Illustrative examples show that the system is capable of recovering instances with high accuracy from very noisy images where standard HT variations falter.