Iterative Boundaries Implicit Identification for Superpixels Segmentation: A Real-Time Approach

Abstract

Superpixel algorithms group visually coherent pixels and form an alternative representation of the regular structure of the pixel grid. This fundamental low-level computer vision preprocessing step greatly reduces the complexity of subsequent image processing tasks. However, most of the existing methods suffer from very high calculation costs which makes them quite unsuitable for time-sensitive applications. In this paper, we propose a new superpixel segmentation method, named IBIS for Iterative Boundaries implicit Identification for superpixels segmentation, that implicitly identifies the boundaries between superpixels and performs the segmentation using only a fraction of the pixels of the input image, thereby reducing the complexity and computation time. The results obtained during the experiments show that the segmentation quality of IBIS is comparable to that of state of the art methods with a computation time divided by a factor of 8 without parallelization of the processing for low resolution images (e.g., 320x240 pixels) as usually provided in public data sets. We also present and comprehensively evaluate the GPU variant of IBIS named IBIScuda that allows an optimal exploitation of the available resources considering the limited bandwidth between CPU and GPU memories.

Medical Image Segmentation Using a Combination of Lattice Boltzmann Method and Fuzzy Clustering Based on GPU CUDA Parallel Processing

“…The author proposes an optimal iteration method to reduce the computation time further. This optimization allowed the authors to accelerate segmentation by a factor of 4 on the Jetson TX1 embedded GPU platform [34].…”

A survey on the state-of-the-art superpixel segmentation

“…We selected a total of 20 superpixel methods, where 13 are recent proposals not evaluated in previous benchmarks: DISF [20], RSS [24], ODISF [25], IBIS [22], DRW [26], DAL-HERS [27], LNSNet [28], ISF [29], SNIC [30], SH [21], GMMSP [31], LSC [32], and SCALP [33]. We also selected the 6 methods recommended in [8]: SLIC [23], SEEDS [34], ERS [35], ETPS [36], CRS [37], and ERGC [38].…”

“…Finally, the Extended Complex Scene Saliency Dataset (ECSSD) [19] has 1000 images with objects and backgrounds with complex textures. Specifically, DISF [20] and SH [21] are recent superpixel methods focused on object delineation, while IBIS [22] and SLIC [23] present high compactness with fair delineation. Finally, we consider a grid-based segmentation (GRID), representing maximum compactness but poor delineation.…”

A survey of superpixel methods and their applications

“…the superpixel boundaries. Only pixels on the borders of these blocks are used to perform segmentation, greatly reducing computational cost Bobbia et al (2021). In addition to SLIC and its variants, some other clustering methods have been proposed.…”