Genetic algorithm matched filter optimization for automated detection of blood vessels from digital retinal images

Due to the importance of the matched filter in the automated detection of blood vessels in digital retinal images, improving its response is highly desirable. This filter may vary in many ways depending on the parameters that govern its response. In this paper, new parameters to optimize the sensitivity of the matched filter are found using genetic algorithms on the test set of the DRIVE databases. The area under the receiver operating curve (ROC) is used as a fitness function for the genetic algorithm. To evaluate the improved matched filter, the maximum average accuracy (MAA) is calculated to be 0.9422 and the average area under ROC is 0.9582.