Image Retrieval Based On Multi-feature Fusion

With the rapid development of remote sensing technology, a large number of image information increases rapidly. Recentlly, a series of satellites launch successfully, which makes more image information providing service for people’s life and production possible.Nowdays image data increasing at an alarming rate.But traditional manual method is time-consuming and outdated, can’t meet the requirement of work. The increasing data makes fast and automatic information extraction method meaningful.Land use(LU) and distribution is the focus of planning management department, and closely related to urban planning and ecological environment protection.Today, the town land is divided into different functional areas according to different use.however, few studies focus on the functional area. Here we choose a typical functional area--residential area(RA) as the study object. RA is closely related to people’s life, and it occupies certain proportion in LU.So it can reflect the status quo of LU to a certain extent. Therefore, the information extraction of RA is of great significance.In this paper, we use image retrieval strategy to extract RA based on multi-feature fusion.We focused on four types of RA(including old village RA, low density high-rise RA, medium density low-rise RA and low density low-rise RA) and defined them respectively. We choose Beijing and Tianjin as research regions, and the classification scheme was build according to the regions. We collected samples with size of 300m*300m extracted from 1m spatial resolution optical image through Google Earth. Reasonable algorithms were designed to extract RA. Firstly, we extract RA based on texture features and fuzzy classification theory.Experiments were carried out to realize the classification between RA and non-RA, and different types of RA. SIFT was introduced to build Bo VW model, and we also improved the traditional Bo VW algorithm. Finally, we proposed an algorithm of fusion classification based on the first two approaches.This algorithm improved the accuracy and efficiency of the classification effectively. The main difficulty lay in discriminating residential areas from mixed categories. The technological process and experimental design in the algorithm can provide a reference for the research of related fields.