Application Research On SVM And Extended Algorithms For Digital Image Processing

Image is an important source of information. Digital image processing has become a focus of domestic and foreign research. As an important part of image processing, image segmentation has attracted a large number of researcher attentions. Because there are no general standards in image segmentation, therefore, most of the traditional segmentation methods are applicable to particular applications. Support Vector Machine (SVM) algorithm for image segmentation is proved to be a more universal approach and has got a good segmentation effect. However, because of SVM has massive computation, when it is used for large-scale data processing such as image segmentation, it has a high computational complexity. Therefore, how to reduce the computational complexity of the SVM algorithm has became a key point of SVM applied to image processing.In this paper, for the Quadratic Programming (QP) problem of SVM, we use the Core Vector Machine (CVM) and Ball Vector Machine (BVM) for the large-scale data learning to verify the effectiveness and advantages of these two algorithms. Both algorithms use computational geometric algorithms to solving QP problems which can significantly reduce the computational complexity. The experimental results show that both CVM and BVM have a similar learning error rate compared to the standard SVM for large-scale data learning. But the time complexity and space complexity of the algorithm in the training process is much lesser than the standard SVM. As an extended algorithm of CVM, the capability of BVM outperforms CVM. Therefore, BVM is more suitable for the large-scale data learning.In image segmentation, the more comprehensive of the image characteristics is, the better of the segmentation effect will be. In this paper, we use the method of moving a 5*5 window on the sampling image to extract the features respectively. The extracted features include the pixel value, the neighborhood statistical features and the texture features of the whole window. Because the extracted feature value is large, so we use the BVM for image segmenting. The experiments show that when using the BVM for image segmentation, the segmentation effect and anti-noise feature of BVM is similar to the standard SVM. But the computational complexity of BVM is lesser than the standard SVM, and it takes lesser time for training. Therefore, when using the BVM for image segmentation, it can avoid the disadvantage of slow training existed in SVM, and it can significantly improve the overall performance of image segmentation.