Research On Algorithm Of Content-based Image Retrieval

With the rapid development of Internet and database technology, the amount of digitalimage information is increased rapidly. Therefore the rapid and effective management andretrieve of valuable information become the urgent needs of people, so the content-basedimage retrieval (CBIR) technology arises at the historic moment, and it becomes the one ofthe very popular currently and challenging research topics gradually. Because the traditionalCBIR technology has certain limitation, namely, its algorithm of image retrieval only extractsthe single feature of image, but the single feature of image has many deficiencies, so thispaper put forward an algorithm of image retrieval based on color and shape features withdifferent weights, and then this algorithm establishes high-dimensional index structure basedon BK-Tree and the improved Fuzzy C-Means (FCM) algorithm for shape feature, finally theretrieval result of this paper is compared with the retrieval results of the traditional CBIRtechnology. In this paper, we mainly explore retrieval algorithm and high-dimensional indexstructure of CBIR technology, the specific research work are as follows:Firstly, in order to make the retrieval results accord with human visual judgment, anddescribe objectively the whole content of image, and improve the retrieval speed and accuracy,this paper put forward an algorithm of image retrieval based on moment invariant andsub-block dominant color (MIDC), which is based on color and shape features with differentweights.Secondly, in view of the disadvantage of traditional FCM algorithm which is that thenumber of cluster must be given in advance, an improved FCM algorithm is proposed, whichcan make the retrieval result achieve or close to the global optimal solution, and greatlyreduce the number of iterations, and make the algorithm faster convergence. And then themoment invariant is established high-dimensional index structure based on BK-Tree and theimproved FCM algorithm (BF), this index structure is not only benefit for reducing the comparing times of the invariant moment of the sample image with the clustering center set,but also FCM algorithm can similar images are classified in same class, which can improveaccuracy of retrieval result.Finally, this paper make a large number of relevant experiments and result analyses forMIDC image retrieval algorithm and BF high-dimensional index structure from theperspective of users. Experiments show that whether the color feature and shape feature of thesample image are obvious or not, the integrated retrieval performance of MIDC algorithm hashigher accuracy and robustness than other traditional algorithms; And experiments also showthat the algorithm based on BF high-dimensional index structure is more preponderant thanthe traditional FCM algorithm both on retrieval time or precision ratio and recall ratio.