This paper describes a novel VLSI architecture for the region-merging algorithm for image segmentation applications. This algorithm uses the region adjacency graph (RAG), which represents regions and their edges. The final segmentation provided by the RAG represents localized contours or surfaces. The architecture is proposed by making use of the concepts of parallelism and pipelining in order to improve the performance in terms of speed. The architecture has been coded in Verilog and synthesized using Synplify tools for FPGA implementation.