Multiple Kernel Learning for Image Contour Extraction

Nowadays, it rapidly increases that digital image information that stored locally or on internet, it has been an important issue that how to save and retrieve images in an effective way. Traditional manual classification can’t meet the actual needs. Now the method commonly used is to extract the image contour, it does not use the full image to deal with the image effectively. Currently, the efficiency of many traditional image contour extraction methods are low for large-scale, heterogeneous digital image, and the result is always poor. For example, the gradient-based method is suitable for straight contour, and the method of prior knowledge drops into the local extreme value easily. According to this, this paper presents an image contour extraction method based on multiple kernels learning (MKL). The method adopts multiple kernels to construct a framework for learning classifier, and then it uses the accumulation of learning classifier and regulates the probability of the output, then it can improve the test results by operator. The experimental results show that this method for contour extraction is more effective. In terms of number, size and accuracy of data collection, this method is much closer to human intuitive initial conditions comparing with other methods, and better effect