Research On Image Classification Based On Semi - Supervised Support Vector Machine

Support vector machine is a supervised machine learning method developed under the statistical learning theory which specializes in machine learning rules of small samples. It seeks to get the best results under conditions of limited sample information and solves the common problems such as curse of dimensionality and nonlinear classification in machine learning at the same time. It is considered to be an efficient and has a superior performance classifier. Image classification based on support vector machine has become an important theoretical and technical in image classification problems. However, it is often difficult to get a large number of labeled data which carrying the correct label, and then the efficiency and effectiveness of the trained classifier will be relatively low. Semi-supervised support vector machine has got more attention as it consider combining the idea of semi-supervised learning and support vector machine method, use a small amount of labeled samples and a large number of unlabeled samples to training support vector machines. It is necessary to apply the semi-supervised support vector machine algorithm to the study of image classification.This paper studies the semi-supervised support vector machine and its application in image classification, fully taps the potential and advantages of the algorithm in image classification. Aiming at a series of algorithm problems applied to image classification such as higher complexity, memory overflow, poor stability, low classification accuracy and so on, we proposed two methods of image classification. The specific works of this paper are as follows:1. We proposed meanS3 VM image classification method based on Mean Shift. Semi-Supervised Support Vector Machine using label mean (meanS3VM) for image classification selects a small number of unlabeled instances randomly to train the classifier then the classification accuracy is low, selects a large number of unlabeled instances then algorithm complexity is high; meanwhile, the algorithm important parameters are estimated by experience, always derives much oscillation of the image classification results. In allusion to the above problems, meanS3VM image classification method based on Mean Shift was proposed. This method combined with mean shift, the smoothed image acquired by mean shift was used as original segmented image to reduce diversities image features; an instance in each smoothed area was randomly selected as unlabeled instance to ensure that it carried useful information for classification and had a more efficient classifier; and the parameters value were also investigated and improved, the important parameter was estimated by combining with support vector machine and mean shift results, so that there will be a better and more stable result. The theoretical analysis and experimental results indicate that this method has achieved a higher classification accuracy and time efficiency in image classification, and avoided the oscillation of the results effectively.2. We proposed semi-supervised composite kernel support vector machine image classification method based on adaptive parameters of Mean Shift. When the semi-supervised composite kernel support vector machine constructing cluster kernel, the universal existence problem is higher complexity and does not apply to large-scale image classification. In addition, when using K-means algorithm for image, the parameter k is difficult to estimate. And the time complexity of clustering will increase rapidly with the increasing amount of data. In allusion to the above problems, semi-supervised composite kernel support vector machine image classification method based on adaptive parameters of Mean Shift was proposed. This method combined with mean shift to make a cluster analysis of the pixel to avoid the limitations of K-means algorithm for image; Adaptive Mean Shift parameters by adjusting the cluster distance and variance automatically combined with the image structural features to get the right clustering results, and then avoided the poor stability problem which is due to determine the parameters by experience; At last, constructed Mean Map cluster kernel with Mean Shift image clustering results to enhance the possibility of the same clustering samples belong to the same category by using Mean Map, then constructed a semi-supervised composite kernel function to guide the image classification by support vector machine. Simulation results show that this image cluster method and adaptive strategy can help algorithm obtaining the image clustering information, and the image classification algorithm which is guided by the composite kernel is more efficient and more stable.