Research On Fabric Image Retrieval Based On Multi-feature Fusion And SVM Classification

With the development of Internet technology and the wide application of multimedia technology,people use more and more images to store information in their daily life.In the face of massive image data,how to accurately and quickly obtain image information is an urgent problem to be solved.At present,although the content-based image retrieval technology can be targeted to solve this demand,the retrieval effect is not ideal,and there is still a problem of information asymmetry between the image characteristics at the bottom and the understanding at the top.Therefore,the design of fabric image retrieval algorithm with multi-feature fusion and SVM classification has practical application value and practical significance.In this paper,the basic feature extraction methods are introduced,and the robust feature extraction algorithm is selected.For the extraction of color features,the color moment based on HSV space is selected.The gray co-occurrence matrix is used to extract texture features.The method of Hu invariant moment was used to extract the shape feature.Since a single feature cannot describe an image clearly,this paper improves on the traditional weighted multi-feature fusion and adopts the method of dynamic weighted multi-feature fusion,which dynamically adjusts the corresponding weights of each feature according to the evaluation of the retrieval result given by the user and the feedback information.The experiment on the self-built image library shows that the retrieval method of dynamic multi-feature weight adjustment not only improves the accuracy of image retrieval but also alleviates the burden of users to set the weight by themselves to some extent.In order to narrow the gap between high level semantics and low level features,the classifier is integrated into fabric image retrieval,and a retrieval algorithm based on SVM correlation feedback is proposed based on traditional SVM classification and correlation feedback technology.Its main idea is to first part by choosing the image library image as the training set to train a classifier with another part of the image retrieval performance test as a test set,and then evaluate the retrieval results by the user and feedback to the system,the system according to the feedback information to adjust the retrieval strategy,the experimental results show that the introduction of relevance feedback technology SVMretrieval method to retrieve the effect is good,have certain practical value.Finally,the fabric image retrieval system is designed and implemented.The system is mainly divided into two parts,image retrieval and background management.Ordinary users can query according to their needs in the image retrieval interface,and administrators can edit image library,set algorithm parameters and manage users in the background in addition to the image retrieval authority.In the experiment,the performance of each module is tested,and the results show that the system has high feasibility.