The Research Of Trademark Retrieval Methods Based On Image Invariant Features

As the number of registered Trademark increasing,the trademark infringement cases increased,for the trademark review and audit puts forward a new problem,the maximum under the premise of reducing the occurrence of the infringement case,for the enterprise registered trademark rights protection.Trademark Image Retrieval(TIR),a branch of Content-Based Image Retrieval(CBIR),is playing an important role in multimedia information retrieval.Aim at the problem of traditional trademark retrieval methods' retrieval time is too long and identified mistakenly.This thesis proposes an effective solution for TIR by combining shape description and feature matching,this thesis puts forward an effective method of trademark retrieval MHU-SIFT plus KMEANS-BOF(modified HU invariant moments-Scale-invariant feature transform-KMEANS and Bag of Features).By use the MHU-SIFT feature extraction in the first place,that is the basic of image preprocessing,and then extract features for MHU moment on the global image,and then extract the local feature-SIFT feature descriptor,image feature point set,then use the KMEANS-BOF method deal with the feature points set,get the clustering center of the library of image feature points,form each image of the word module(Pitch),namely the BOF,formation of BOF feature vector library,for image retrieval.Feature matching part using the included Angle cosine to calculate the similarity,and the final output image by the image similarity score.In order to verify the validity of the method,establish a platform in VS2010 for the experimental simulation and invoke MATLAB2015 b,through the standard image MPEG7_CE-Shape-1_Part_B and trademark image library,a large number of test evaluation our proposed method.Theoretical research and experimental results show that in the process of image retrieval in this thesis,the MHU-MEANS-BOF SIFTK fusion method,compared with the traditional HU-SIFT,accuracy and rapidity of the method is validated.The proposed method has a wider application range.