Ellipzoid 1.0: a multiple population genetic algorithm for detection of multiple elliptic objects in noisy microscopic image

点击翻译按钮获取中文摘要

The purpose of this project is to produce an algorithm that can detect ellipses and elliptic curves in binary images.  To be useful for real-world applications, the algorithm must be robust enough to detect malformed ellipses and be able to scan images that have background noise.The algorithm partitions different populations of objects, each of which forms a cluster around a potential elliptic curve. Each object represents a candidate ellipse. The algorithm uses both evolution and clustering to gradually better approximate the actual elliptic curve, presumably at the centre of each population, until that curve is detected and the population is terminated. This process occurs simultaneously within all populations in the image until all elliptic curves are located.We have comprehensive data that show that this algorithm operates faster and more accurately (especially in cases in which images have multiple elliptic curves and/or noise) than the standard technique used for ellipse detection, the Randomized Hough Transform (or RHT). In fact, we were awarded a prize at GECCO 2006 for providing evidence that we have achieved exactly that objective.There are many potential applications for this technology, including the detection and counting of cells in microscopic images, the detection of hand-writing patterns, etc.