Construction of an image descriptor from graph representation based on a primitives detected by FAST

Extraction methods of points of interest in image characterization are many and diverse. In general the process of characterization follows a construction of a detector/descriptor schema. The detection allows selecting some points of interest as primitives who capture important structural information about the image. These detected points are then described by characteristic vectors to form a representative signature of the image. We find many applications of these methods in systems of classification, indexing and search by content in the images databases. While most of detection methods offer descriptors related both to the detection phase and the destination application. For the FAST (Features from Accelerated Segment Test) method, it remains one of the fastest methods currently, and located many application in the tasks of instant characterization. However, the authors of this method offer only a detection approach, without explicitly giving the approach of description to let this part for future application methods. On the other hand, others methods based on graphs, bringing structural aspects over the statistical methods of characterization. In this paper, we propose a method of characterization of the image by the graph that uses points of interest detected by the FAST algorithm to construct the structure of graphs of characterization. Thus the image similarity transforms into a problem of graphs matching. We then describe the advantage offered by this method over conventional characterization approaches.