Research Of SOM-based Contour Extraction In Multiple Objects Image

Contour extraction is one of the most important aspects of image processing and computer vision. However, existed technology of contour extraction could not reflect the targeted information in the image. So the technology of multiple contours extraction for the image which involved multiple targets has been the focused research problems.ACM (active contour models) provide the new paradigm for contour extraction with deformed model based method. After a lot of years'innovation, the ACM break new ground when artificial neural network is involved. The most sophisticated approach is batch-SOM based ACM. This method is extended to handle extraction of contours of multiple objects in image.In this paper, the main work is to improve the BSOM method to extract multiple contours. First, an improved FN-SOM (Self-Organizing Map considering the False Neighboring Neuron) is proposed which changes the neighborhood relationship more flexibly according to the situation and the shape of data comparing with the conventional FN-SOM. Learning performance is evaluated both visually and quantitatively using the three measurements. In a word, we demonstrate the efficiency of algorithms with emulation and analysis.Next, this paper proposes the improved method in two aspects. One of them is to handle the colored image. The procedure is to extract contour with traditional BSOM after the segmentation of colored image using FN-SOM. Although this method could extract multiple contours effectively, the method did not bring essential improvement to conventional BSOM and have the high algorithm complexity.The most important point in this paper is the improvement for the architecture of BSOM. Drawing on the concept of FN-SOM, this paper proposes FN-BSOM based ACM to extract multiple objects'contour. The contrast experiments demonstrate the novel method could extract the contour of complex images in which there are multiple objects involved, more effective then conventional BSOM.