Font Size: a A A

Research Of Multi Feature Fusion Technic Applied To Content Based Image Retrieval

Posted on:2009-12-29Degree:MasterType:Thesis
Country:ChinaCandidate:Z LiuFull Text:PDF
GTID:2178360245454069Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
This dissertation discusses some key problems on content-based image retrieval technique focusing on shortcomings of current systems in efficiency and precision. A multi-feature fusion approach is proposed suitable for such systems. The discussion emphasizes on feature extraction, multi-feature fusion, and fuzzy similarity measure. Based on the discusses above, a experimental system is designed, and its effectiveness is proved in the experiments.Problems studies in the dissertation include: first, object boundaries are often uncertain in most nature-view images. It goes against region-based retrieval. A multi-feature fusion approach is proposed to express which region one pixel belongs to, and it is suitable for the uncertainty. Second, most current systems employ global feature for retrieval, and thus, background which possesses most area of an image determines the result of retrieval, and users are often confused with these results. Global and semi-global feature are combined to represent features in different metric to solve this problem, and a balance is obtained between precision and effectiveness.Main innovations of the dissertation are:1. The EM algorithm is explored to segment an image into several coherent regions, each of which coarsely corresponds to one object in the image. Considering the uncertainty of object boundaries, a multi-feature fusion approach is proposed to express which region one pixel belongs to, and it is suitable for nature-view image retrieval.2. A fuzzy similarity measure is proposed based on fuzzy feature representation. This method makes it possible for any two regions in two images can be measured in feature space so that attention is given to effectiveness of the system in the precondition of precision.
Keywords/Search Tags:Multi-feature fusion, EM algorithm, Semi-global feature, Fuzzy similarity measure
PDF Full Text Request
Related items