Flooding region growing: a new parallel image segmentation model based on membrane computing

Region-growing (RG) algorithm is one of the most common image segmentation methods used for different image processing and machine vision applications. However, this algorithm has two main problems: (1) high computational complexity and the difficulty of its parallel implementation caused by sequential process of adding pixels to regions; (2) low performance of RG in region with weak edges, due to the use of location and the number of seed points. In this paper, a new model of RG algorithm based on tissue-like P system is proposed to resolve these limitations. In this model, each pixel is modeled by a membrane, and in one step, the similarity of each membrane with its neighbors is computed. Then, all membranes are used as seed points to grow simultaneously in a parallel and flood-like manner. To realize the parallel implementation of the proposed model, Graphic Processing Unit (GPU) and CUDA programming language are used. The evaluation of execution time indicates that the proposed model has better performance than the conventional RG algorithm, its speed-up is about 12.5x. Qualitative and quantitative evaluations of segmentation performance also demonstrate that the proposed method not only does not damage the overall segmentation accuracy, but also it has better results on images with complicated background compared to the state-of-the-art methods.

Automated summary

The study introduces a new model of the RG algorithm based on tissue-like P system to address the high computational complexity and low performance issues of the conventional RG algorithm. Through the utilization of GPU and CUDA programming language, the proposed model achieves better performance with a speed-up of about 12.5x. Qualitative and quantitative evaluations demonstrate that the proposed method not only maintains overall segmentation accuracy but also performs better on images with complicated backgrounds.