Novel intelligent edge detector for sonographical images

• Authors: Ali Rafiee;Mohammad Hasan Moradi;Mohammad Reza Farzaneh
• Venue: EC'05 Proceedings of the 6th WSEAS international conference on Evolutionary computing
• Year: 2005

Most image processing, such as image registration, image segmentation, region separation, object description, and recognition, use edge detection as a preprocessing stage. Real ultrasound images, such as sonography images, can be corrupted with speckle noise. The real problem is how to extract the edges and simultaneously preserve image details. In this paper a new genetic-neuro-fuzzy system is suggested for edge detector in ultrasound images. The competitive neural network (NN) is used for this system. Data processing will be done by a winner-take-all competition process is applied to subnetworks in NN and neurons in each subnetwork. The fuzzy transformer system is used to convert the neighborhood window of input pixels to three decision fuzzy parameters. The on-line genetic algorithm (OGA) is used to optimize and regulate the system parameters. A binary pattern of neighborhood window is obtained based on winner subnetwork and neuron. After detecting the first set of edge pixels, next structural algorithm will be applied according to the location of edge pixels to eliminate some of the noisy edges and add some weak real edge pixels. System performance is compared with the standard methods such as Sobel and zero-crossing edge detector. Results show that the genetic-neuro-fuzzy edge detector is a powerful edge detector, whose performance is better than standard edge detectors.