Morphology Based Scale Invariant Feature Transform And Its Application In Image Retrieval

In nowaday society, both of the storage and application of data image develops fast. How to effectively extract and use the effective information of these images is always the core research problem for the scholars at home and abroad. Traditional image index is based on text, it has many defects, which includes hard representing image information, different understanding of same picture by different people, big work amount. So, an image index technique, which is based on image content, and which can extract and index the image content automatically and objectively, has important research meaning and practical value.In this Thesis, we introduce some image feature extraction methods and how these methods can help image retrieval. We introduce image feature extraction methods, image key points extraction methods, and image similarity comparing comparison methods.Through researching the classical local feature extraction algorithm, SIFT algorithm, we find that for two same kind objects which have different texture and gradient, we cannot use SIFT algorithm to detect the matching of the two objects by using eigenvectors of matching key points. To solve this problem, in this thesis, we also propose a new feature extraction method for image retrieval. The method describes the morphology relationship between the SIFT key points whose position is calculated by the SIFT algorithm and the edge of an image. We use the length between the SIFT key points and the edge of an image, as well as open and close relationship to describe the morphology relationship. On the basis that we extract the key points through describing the key points, we propose the definition of key points dissimilarity, this method compares the location of the key points as well as compares the description of key points, in order to better match the key points of different images. To match two different images, we first match key points of two images and we calculate the dissimilarity value of one image by comparing with another, then we add the normalized dissimilarity value of two images as the total dissimilarity value and use it to retrieval image. We call the proposed feature extraction algorithm as Morphology-based Scale Invariant Feature Transform Algorithm.The thesis introduces the background of the proposed image retrieval method, including the SIFT algorithm. The thesis also does a lot of experiments on the proposed image retrieval algorithm and compares the algorithm with some other proposed image retrieval methods which is written by other people and similar to our proposed algorithm. The experiment result shows that our proposed method can effectively describe the image feature and can be effectively used in image retrieval process. In the end, the thesis proposes some future works which may improve the proposed algorithm.