A Cascaded Retrieval Method of Specified Object Based on Fusing Multiple Features

Considering the need of video retrieval and images retrieval, a cascaded retrieval method of specified object based on fusing multiple features has been put forward in this paper. The color and the SURF features should be fused subtly by the algorithms. And then the retrieval object can be extracted gradually from the rough to the subtle by the cascaded retrieval method. It mainly includes the following 3 parts: rough retrieval on the first level, subtle retrieval on the second level, comprehensive judgment on the third level. The rough area containing the suspected vehicles can be recognized by the color histogram with color layered in rough retrieval. Further recognition on the basis of rough retrieval can be done by obtaining SIFT features information in subtle retrieval. The comprehensive confidence of the object vehicles can be obtained by some principles according the independent similarity from the rough retrieval of colors and the subtle retrieval of features in comprehensive judgment. The experiments show the problem that the object features cannot be extracted from rigid object and flexible object at the same time has been solved by the algorithm in this paper. It has a better retrieval precision and a high processing speed, so that it prove to a new method to retrieve the video and image effectively.