Research And Application On Invariant Feature Points Of Mms Image Retrieval

With the rapid popularization of mobile phone, operators are cultivating customers'habits of sending MMS through rich MMS provided by the MMS content providers (CP), promoting the development of point to point MMS meanwhile. To enhance the control of MMS, mobile operators need to distinguish issued MMS which are provided by the CP. Only in this way they can be introduced the content list, copyrighted content forwarding, which is necessary to promote high-quality MMS delivery and increase the number of point to point MMS business and utilization. This requires usage of content-based image retrieval techniques (CBIR) for the retrieve of the MMS provided by content providers.Invariant features detection is an important theoretical basis for image retrieval. The SIFT descriptor has been proved to be the most effective local invariant descriptor for practical uses nowadays. Experiments show that SIFT algorithm is invariant to scale transformation, illumination changes, rotation, noise and other factors. However, like most algorithms, the SIFT algorithm's matching happens after the color image is transformed into gray one, which causes color information loss and may lead to wrong matching. In addition, the high dimensionality of the SIFT descriptor needs large computation. This paper researched the existed image matching techniques of invariant features and gives a MMS matching system framework based on invariant features.To solve the problem of low efficiency of high-dimensional space search, this paper proposed an improved algorithm based on k-d tree search, which can find the nearest neighbor points with high probability and less search time. To solve the problem of time-consuming in feature extraction phase, this paper proposed a MMS matching system based on SURF(Speed Up Robust Features) descriptor. The algorithm which is based on the Hessain matrix, describes a distribution of Haar-wavelet responses within the interest point neighborhood and relies on integral images to reduce the computation time. In order to make full use of the image color information, this paper uses the color invariant theory to enhance the robustness and extraction efficiency of the descriptors, and then uses the fast index method to reduce match time, resulting in the similarity matching has been improved on.Finally, the paper analyzed three kinds of algorithms in the experiment. The experimental results show that SURF and its improved algorithm have a good performance and robustness compared to SIFT descriptor. It fully suit the MMS image matching system.