Research Based on an Active Relevance Feedback Mechanism in Content-Based Image Retrieval
|School||Northeast Normal University|
|Course||Applied Computer Technology|
|Keywords||Content-based image retrieval system User interest Logistic regression model Active Learning Relevance feedback|
In this paper, based on a logistic regression model ( logistic regression , LR ) feedback mechanism to effectively improve the image of low-level visual features and high - level semantic features divide , and ultimately improve the content-based image retrieval ( CBIR ) the efficiency and accuracy. This article will interest the user preferences which added to the feedback mechanism , user interest modeling , the formation of a probability distribution , and ultimately the use of this model to predict the image database belonging to the user interest categories, the probability value , and in accordance with the value of this probability returns the result image collection of the sort output . This process is so constantly repeated, until the user is satisfied with the query results or until you find the target image . Relevance feedback will encounter a critical issue , that is, user feedback, the number of samples , that is, the number of training samples is much smaller than the image dimension of feature vectors , the logistic regression model can not adjust the overall parameters , this article from two aspects to solve this problem. First of all , the use of iterative logistic regression model (Iteration Logistic Regression, ILR) method , the original image feature vector space into several small subset of the first model in each subset of the internal , and then trained as a subset of the eigenvectors of the internal vector modeling . In addition, the use of machine learning of the more popular active learning algorithm ( Active Learning , AL ) , pick out those with the greatest amount of information unlabeled samples to allow the user to mark not only expanded the number of training samples , the most important thing is to pick out the classification advantageously a sample so that a user mark , and optimization of the classification results . Experimental results show that the proposed scheme can improve the efficiency of the image retrieval system .