The architecture of a software system is often described with diagrams embedded in the documentation. However, these diagrams are normally stored and shared as images, losing track of model-level architectural information and refraining software engineers from working on the architectural model later on. In this context, tools able to extract architectural information from images can be of great help. In this article, we present a framework called IMEAV for processing architectural diagrams (based on speci c viewtypes) and recovering information from them. We have instantiated our framework to analyze \\module views and evaluated this prototype with an image dataset. Results have been encouraging, showing a good accuracy for recognizing modules, relations and textual features.