Image Retrieval Based On Geometric Partitioning For Edge Structure

This paper provides a new image retrieval approach: Image retrieval based on geometric partitioning for edge structure. This approach represents image features by "Edge-Pixel Segment (EPS)", and establishes the index using mature text-based information retrieval technology after block coding to edge image.The spirit about this image retrieval approach derives from text-based IR. There are two major problems to settle in order to generalize text-based IR to image retrieval. One is what are "keywords" of an image and how to generate them, the other is how to establish the index using the text-based indexing models in the image domain."Edge-pixel Segment (EPS)" proposed in this paper is an important concept to represent edge structure features. It is generated by means of geometric partitioning for edge image and described by eigenvector. This new image representation method has many advantages. It can reduce gray-level redundancy brought by partitioning gray-level image directly, and can avoid the difficulty of image segmentation and object-oriented descriptions semantically. It is important that it has invariance of rotation and translation.Clustering the eigenvectors of edge images in training set to generate the codebook, "KeyEPSs", encoding the edge images in image database by vector quantization, the images in database and the retrieval image can be represented by one dimension data, which is similar to the text data structure. At this time, text-based IR technology can be used to image retrieval. Through Boolean Model of inverted file and Histogram Model, a new index is established aimed at syntactically representation of image structure content.