DATA STREAM PROCESSING ARCHITECTURE ENABLING EXTENSION OF NEIGHBORHOOD MASK

摘要

The invention relates to a data stream processing architecture forming NI x Ne pixel matrices. The processing architecture (20) comprises processing units (UT) and an interconnection means (MI) enabling the transfer of the data stream between said processing units. Each processing unit (UT) comprises a control unit (UC), neighborhood processors (PV) that are capable of carrying out the commands from the control unit (UC), and a neighborhood storage unit (UMV) enabling each neighborhood processor (PV) to access a VI x Vc neighborhood. The invention enables the size of the neighborhoods accessible by the neighborhood processors (PV) to be extended, by coupling two or more processing units (UT). The architecture (20) comprises, to this effect, a storage manager (GM), enabling the neighborhood processors (PV) of the coupled processing units (UT) to access adjoining pixel neighborhoods, and moreover comprises a means for communication between neighborhood processors (PV), thus enabling an exchange of data related to the various neighborhoods.