Image representation system for semantic context description

In this document we disclose an architectural solution for obtaining an image representation model which can be used for to semantically summarize the context of the scene. This image representation model can be exploited in many application contexts, such as semantic image categorization/recognition, image retrieval, and image clustering. Images (or frame of a video) are holistically represented starting from the statistics collected on DCT domain. Since DCT coefficients are usually computed within a digital signal processor for the JPEG or MPEG conversion, the proposed solution allows instant and “free of charge” image representation process. This means that the proposed representation model can be derived at very low cost in mobile devices, such as smartphone, tablets, etc. and in general in all IGP in which the DCT is included.