Object-based Image Retrieval

Currently, content-based image retrieval is the research hotspot. Effectively using vision features in an image is necessary parts to retrieve image. But ,due to mostly used image vision features in content-based image retrieval are overall , local information does not be emphasized(embodied) and prominent parts . On the other hand, local intact parts can't be extracted so that user partial need can't be satisfied. The basic thought in this article is to segment an image into some relative homogeneous regions which in some degree represent the objects (sub-objects) contained an image. Based on the regions segmented we define the features of regions and similarity which is used to image retrieval. The selection of seed pixels and the number of regions is the research center. A lot of experiments show that the threshold acts on a important role to determine the number of sub-block and to show out the detail parts. The definition of local pixel difference suits to segment an image effectively and based on segmentation, the method can support the single object inquiry.