Image Retrieval Algorithm Design Of Relevance Feedback Based-on Support Vector Machine

Machine learning learns the low-level features of the sample image set and classifies it according to the machine learning mechanism,thus obtaining more accurate search results.Support Vector Machine(SVM)is a machine learning method.Its essence is a twoclassifier.It can convert the linear indivisible problem of low-dimensional space into linear separability of high-dimensional feature space by selecting appropriate kernel function.That is to solve the extremum problem of the objective function under the constraint condition,thus effectively avoiding the local optimal solution problem in the artificial neural network such as BP(back propagation)and other neural network classification methods,so as to obtain better retrieval effect..However,the retrieval subject of contentbased image retrieval is more inclined to computers,and the retrieval user and the retrieval system are not able to interact,so that the obtained retrieval results can not fully meet the user requirements.Therefore,the user retrieval behavior module,that is,the related feedback,is added to the retrieval system,and the user selects multiple images for continuous retrieval.Content-based image retrieval(CBIR)is mainly based on extracting low-level features such as color features,texture features,shape features and spatial features,using similarity measurement methods such as Euclidean distance and Mahalanobis distance.The matching between the search image and the search image library is performed,and the images with similar similarity are arranged and displayed,and the precision and the recall ratio can be used as the evaluation method of the search performance.In this thesis,a correlation feedback image retrieval algorithm is designed.By searching the user,the interested and non-interested images are selected as training samples in the first retrieval result,and the related feedback algorithm is used to learn,so as to realize the prediction of the image library.The difference between the retrieval accuracy of the single feature and the comprehensive feature of the comparison image,the difference of the retrieval results of the different similarity measurement methods,the difference of the retrieval results of different kernel functions,and the difference of the retrieval results of different classifiers.Support vectormachine is used to combine image color features,texture features,image segmentation features,K-means clustering algorithm and word bag model for image retrieval,and the design of related feedback image retrieval system is completed.By constructing multiscale wavelet kernel support vector machine and introducing active learning mechanism and bulldozer distance measurement method,the retrieval results of different feedback times are compared,and the active learning and unused learning retrieval results are used.