An image indexing method is presented to perform the CBIR (content based image retrieval) for about 100 different images. The method is based on the fractal framework of the PIFS (partitioned iterated function systems) widely used for image compression. The image indexing is represented through i-model features. A specific searching algorithm allows to use this representation as an indexing to retrieval an images. To optimize the matching and retrieval process, an indexed data structure is used to store the PIFS and to calculate the distance between the compressed images and the index image. Also, we present the theoretical arguments to justify their use in image retrieval.