Multiple scale image segmentation-based multiple scale approximation with support vector machines

Abstract

Aiming at the low accuracy of wavelet multiple scale segmentation with current common algorithms, one multiple scale approximation segmentation with support vector machines SVM-MSA is proposed in this paper. According to the proposed segmentation algorithm, the results of decomposed images at deep levels by MSA would be further segmented, which is actually mapped to the decomposed images at last levels. Then, each region from last segmentation would be further segmented according to the segmentation algorithm, and thus the segmented images with finer results. So, the nesting image segmentation could be finally obtained. Different from the classical multiple scale image segmentation algorithms, the information at the same location with different scales could be avoided to be segmented and the segmentation from fine to coarse scales could also be avoided without any amalgamation of coarse segmentation. Results of numerical evaluations showed that the consistent segmentation results could be obtained to that of human vision analysis without over segmentation of image contents.