A new method of training samples choose for support vector machines

Choose of training samples is an important topic for the support vector machines applied research. This paper proposes a new image segmentation method: the two-dimensional histogram of the segmented image is used to guide to select training samples to train the classical support vector machines. First of all, calculate the two-dimensional histogram of the segmented image and then get its two thresholds by the rough estimation which proposes in the paper; then, according to these two thresholds to find the range to choices the training samples; finally, choose the appropriate number of training samples in the certain range to train support vector machines. The method have better results whether in visual or in quantitative analysis.