Interactive Image Retrieval Based On PSO Weight Vector

In recent years, with the rapid development of computer and communication technology,multimedia information appeared a rapid growth trend. the amount and type of digital image were alsogrowing, how to quickly extract visual information from a large number of images, had became an urgentproblem in the information age for people, classification and retrieval of image database became one of thehot spots to get image information. Content-based image retrieval had been widespread concern, which wasmainly retrieved by the underlying feature extraction image (texture, shape, color) to achieve similaritymatching between images.Along with the development of swarm intelligence techniques, particle swarm optimizationalgorithm application in various fields has made great progress and developments, such swarm intelligenceoptimization algorithm was different from the traditional optimization algorithms for nonlinear,multi-extreme characteristics complex function and combinatorial optimization problems provide practicalsolutions.However, because the underlying characteristics of the image were currently no ability to fullyidentify the objects contained in the image, and how people’s subjective opinions added to the imageretrieval was an important research focus for CBIR, with the development of the information retrievaltechnology, relevant feedback technology was introduced into the image retrieval in the1990s, which toreduce the gap between the high-level semantic concept with the same underlying characteristics providedan effective way.In this paper, applying the particle swarm optimization algorithm combined with the relatedfeedback, based on the weight vector, two content-based image retrieval methods are proposed: One is based on the color histogram of the weight vector associated feedback: Positive correlationfeedback framework based on the PSO optimization color histogram of the weight vector、the positive andnegative relevance feedback framework based on the PSO optimization color histogram of the weightvector; based on transformed weight eigenvector. We first determine the user’s initial inquiry to point themovement formula, according to user feedback on the results of the previous output to adjust the weightsquery vector, let the weight inquiry vector away from the counter-examples and close positive examples,on the basis get the further optimization of the weight inquiry vector, which will be more close to the realintentions of the user.The second is associated feedback based on the integration of feature weight vector: The positivecorrelation feedback framework based on PSO optimizing the integration of features weight vector、thepositive and negative relevance feedback framework by the PSO optimization the integration of featuresweight vector. Contrasting a single feature only can express some properties of an image, by combining thefeature vector of textures, shapes and colors, we can get a comprehensive weight vector of the textures,shapes and colors, and using particle swarm optimization (PSO) algorithm we can optimize weight queryvector and update it.Finally, contrasting the natural image with Corel1000by experiment, and comparing relatedalgorithms with our previous search algorithm performance, it can verify the effectiveness of our algorithm,and the weight frame for natural search has ideal results, and possesses certain advantages.