An improved image segmentation algorithm using support vector machines

Abstract

An improved image segmentation algorithm based on support vector machines was proposed, which belonged to hybrid segmentation techniques. Considering image segmentation based on support vector machines required the user to provide the training data, an automatic data providing method was proposed to obtain training data used by support vector machines instead of directly taking some pieces of the image by user. In the improved algorithm, feature vectors of homogeneous region were firstly classified using unsupervised classification technique, and then feature vectors and class labels were fed into support vector machines for training and latter for predicting the labels of unknown samples once the training was complete. The experiments show that the proposed algorithm is efficient for both smooth image segmentation and texture image segmentation. Meanwhile, the classified model trained using one representative image can be applied to the set of similar images and 3D volume data.