Flower Image Retrieval Based On Memetic Feature Selection

With the rapid development of technology, digital image resource has grown explosively. It's become more and more vital to provide effective image retrieval methods to help people to get information they wanted. We conduct a systematic and deep study on flower image retrieval problem and proposed a novel flower image retrieval method based on feature selection using Memetic algorithm.The main work and contribution of this thesis are listed as follows:(1) Introducing the development trend and some key techniques of content-based image retrieval(CBIR). And giving an introduction of the flower image retrieval problem.(2) Using a Memetic framework embedded a local search operation named ‘Markova blanket' to select optimal feature subset. We used two operations named “add” and “delete”. With this two operations, we delete a lot of redundant and irrelevant features and improve retrieval efficiency at the same time.(3) Improving the hierarchical classification method based on multi-feature fusion to solving the inter-class similarity and intra-class difference issue. We use Memetic algorithm to obtain optimal subset by training a classifier for each type individually, obtaining probability of each category, and combining those classifiers.(4) Using Oxford 17 dataset and Corel dataset to test our method. The experimental results shows that the proposed method overcomes the intra-class difference problem and achieves effective retrieval results.