Font Size: a A A

SVM Algorithm Study And Its Application In Object Detection

Posted on:2018-10-25Degree:MasterType:Thesis
Country:ChinaCandidate:S W SunFull Text:PDF
GTID:2348330515987168Subject:Electronics and Communications Engineering
Abstract/Summary:PDF Full Text Request
Machine Learning(ML)is an important subfield of computer science.Arthur Samuel,gives it definition--"computers with the ability to learn without being explicitly programmed." That is to say,ML,to a certain extent,gives the computer an ability of"thinking".Machine learning has a close relationship with computational statistics,mathematical optimization and data mining,while there are some differences at the same time.ML can be divided into three categories:supervised learning,unsupervised learning and enhanced learning,according to whether there is feedback in the learning process or not.Support vector machine(SVM)is a supervised learning algorithm that can be used to solve classification and regression problems.SVM deals with classification or regression problems by constructing hyperplane in the feature space,and the model with ability to handle linear situation is extended to the nonlinear case by the kernel trick.SVM is one of the best ML algorithms and can be used to solve many problems in practical applications.SVM has been widely used in text and hypertext classification,image classification,handwriting recognition,bio-metrics recognition etc.The parameters of the kernel function and other related parameters of SVM are usually set based on experience,and the final performance of the model usually depends on the selection of parameters.In this paper,particle swarm optimization algorithm and artificial bee colony algorithm are used to optimize the selection of SVM parameters.Compared with other SVM parameter selection methods,the SVM model based on the group intelligent optimization algorithm can obtain better generalization performance,since the group intelligent optimization algorithm has the advantages of requiring no continuous function and having the ability to jump out of local extrema.Only over ten years ago,using machines to recognize object and retrieve image seemed to be an impossible task.With the popularity of the Internet in recent years,more and more images appear on the Internet,massive image data makes the artificial image processing and recognition impossible to achieve.Fortunately,researchers have been committed to researching the computer vision technology,giving machine the ability to deal with image recognition,classification,retrieval and other tasks instead of doing them artificially.SVM algorithm shows excellent performance in the process of image recognition,classification,retrieval and other computer vision tasks.Exemplar-SVM is a recently proposed linear SVM model trained with a single positive exemplar and a negative exemplar set.The algorithm has been applied in the field of object detection,content-based image retrieval(CBIR)and so on.E-SVM trains a linear SVM classifier for each positive exemplar,and we finally get a set of E-SVM models.Experiment on the PASCAL VOC 2007 object detection data set shows that the performance of E-SVM is on a par with Latent Deformable Part Model which is much more complex.E-SVM model trains each sample set to get a number of specific detectors,in this paper,we propose to use K-means to cluster the E-SVM detectors,obtaining a set of detectors with average feature of the samples.On one hand,the resulting detectors have a better generalization performance because of the integration of multiple features of the corresponding object.On the other hand,the number of detectors after clustering is greatly reduced,thus the recognition time can be reduced and the detection efficiency can be improved during the object detection.
Keywords/Search Tags:Support Vector Machine, Particle Swarm Optimization, Artificial Bee Colony, Image Feature Extraction, Object Detection, K-means, E-SVM
PDF Full Text Request
Related items