Content based image retrieval using textual representation of SIFT features
MetadataShow full item record
This paper presents a new method for SIFT feature image matching that can be readily used and applied in a standard search engine. For any object there are many features, interesting points in an image, that can be extracted to provide a feature description of the object. The SIFT approach for image feature generation takes an image and transforms it into a large collection of local feature vectors, which make image retrieval possible. In order to match an image within a large database of images, a match for a given feaure is traditionally defined as the local feature vector with minimum Euclidean distance from the first. As this method does not transcribe to search engines, a new method was developed that allows for the quick retrieval of images. It was found that only a limited number of indices, the ones with the largest values, in any feature vector need be analyzed when finding reliable matches with a search engine. Using a search application, this method was tested and refined. Real-time search performance was achieved.