Dissertation > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device

Research for CBIR Based on Multi-intelligent Algorithms and Image Fusion

Author FuQiMing
Tutor LiuQuan
School Suzhou University
Course Management Science and Engineering
Keywords CBIR reinforcement learning genetic algorithm cluster image fusion relevance feedback
CLC TP391.41
Type Master's thesis
Year 2011
Downloads 46
Quotes 0
Download Dissertation

Along with the development of the computer, network and multimedia technology, the number of the digital image increases exponentially, so how to retrieval images from mass image data is a big problem today. The traditional image retrieval methods use manual text annotation for images, and then retrieve images using key words. But the biggest drawback is time-consuming and large workload for manual text annotation, and the text annotation depends on the subjective judgment of marked people to a large extend, which is not conducive to the objective description. In this background, the CBIR (Content-Based Image Retrieval) was proposed, which is based on the content of images, extract image features and retrieval images with features. At present, many domestic and foreign institutions are doing the relevant research and get many achievements, but the low retrieval efficiency and precision is still a big problem, which restricts the application of this method.In allusion to the problem of the low retrieval precision and efficiency, combining the knowledge of image processing technology and machine learning, this thesis put forward a series of improvement methods, which mainly include the following five aspects:(1) In allusion to the problem that single feature can’t describe image accurately, based on the image color, texture and shape, put forward an image retrieval method with multi-features. The experiments show that the method has higher retrieval precision compared to the method with single feature.(2) In allusion to the problem that how to enactment the weights for features, combining the genetic algorithm, put forward a novel image retrieval method based on relevance feedback and genetic algorithm, which can enactment the weights automatically in order to modify the similarity model. The experiments show that the method has better performance on retrieval precision and efficiency.(3) In allusion to the problem that the normal vector correction formula can’t reflect user’s needs correctly, put forward a novel retrieval method based on genetic algorithm and image fusion, which can modify the query vector according to the feedback by using image fusion technology. The experiments show that the method has higher retrieval efficiency with the guarantee of retrieval precision.(4) Separate the image library and feature library in the retrieval process, and put forward a novel K-Means cluster algorithm based on genetic algorithm, which can do image classification, extract representative features and improve the retrieval efficiency.(5) Based on reinforcement learning, combining genetic algorithm, cluster algorithm and image fusion technology, construct an intelligent retrieval framework with active learning ability for CBIR, and carry out intelligent retrieval eventually.

Related Dissertations
More Dissertations