Content-based image retrieval is the task of recalling images from a large database that are similar to a probe image. Many schemes have been proposed and often follow the scheme of extracting information from images and classifying this information as a single entity. We propose that image segments are far more complicated and that two adjustments are necessary. The first is that pixels do not necessarily belong to a single object and the second is that image segments can not be classified as a single entity. We propose a new approach that adopts these tenets and present results indicating the feasibility of creating syntactical definitions to image objects.