A Fast Parallel Implementation of Queue-based Morphological Reconstruction using GPUs

In this paper we develop and experimentally evaluate a novel GPU-based implementation of the morphological reconstruction operation. This operation is commonly used in the segmentation and feature computation steps of image analysis pipelines, and often serves as a component in other image processing operations. Our implementation builds on a fast hybrid CPU algorithm, which employs a queue structure for efficient execution, and is the first GPU-enabled version of the queue-based hybrid algorithm. We evaluate our implementation using state-of-the-art GPU accelerators and images obtained by high resolution microscopy scanners from whole tissue slides. The experimental results show that our GPU version achieves up to 20× speedup as compared to the sequential CPU version. Additionally, our implementation’s performance is superior to the previously published GPU-based morphological reconstruction, which is built on top of slower baseline version of the operation.