ASSIMIV - Detection and description of geographical objects by matrix and vector information assimilation

Context and objectives

This study focuses on an advanced method to complete, enrich and update a large scale vector database using VHR images. Existing vector information combined with VHR image-segments generates additional contextual and dynamic information beside the features obtained from image or vector separately. The originality of this research is to propose a robust classification scheme where spectrally irrelevant pixels are first concealed and then reintroduced as meaningful contextual information. It aims to implement an operational workflow to obtain a good positional and thematic accuracy using a single VHR image and an existing vector database.