Abstract: A new line extraction accelerating algorithm based upon graphics processing units (GPUs) is presented. Several techniques are proposed to optimize the GPU computation. Among those improvements, the most important one is the use of "seed pixels", which allows us to extract the lines while benefiting from the parallelism GPU architectures offer. Only a few bytes of data are transferred between GPU and CPU. The experimental results show that the proposed algorithm has a significant increase in speed when compared with a stand-alone CPU. In comparison to a traditional CPU implementation, the performance gains range from x5 to x10, depending on the image size.