SEEDS: Superpixels Extracted Via Energy-Driven Sampling

By Michel van den Bergh, Xavier Boix, Gemma Roig, Benjamin de Capitani, Luc Van Gool. 2012

747 citations recorded by the source corpus.

Superpixel algorithms aim to over-segment the image by grouping pixels that belong to the same object. Many state-of-the-art superpixel algorithms rely on minimizing objective functions to enforce color homogeneity. The optimization is accomplished by sophisticated methods that progressively build the superpixels, typically by adding cuts or growing superpixels. As a result, they are computationally too expensive for real-time applications. We introduce a new approach based on a simple hill-climbing optimization. Starting from an initial superpixel partitioning, it continuously refines the superpixels by modifying the boundaries. We define a robust and fast to evaluate energy function, based on enforcing color similarity between the boundaries and the superpixel color histogram. In a series of experiments, we show that we achieve an excellent compromise between accuracy and efficiency. We are able to achieve a performance comparable to the state-of-the-art, but in real-time on a single Intel i7 CPU at 2.8 GHz.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

What Makes for Effective Detection Proposals? (2015, background): “The authors show that using multiple superpixel maps signiﬁcantly improves recall.”
Creating Summaries from User Videos (2014, methodology): “We term these segments superframes, in analogy to superpixels, and propose an approach inspired by recent work in image segmentation [2].”
ViewAL: Active Learning With Viewpoint Entropy for Semantic Segmentation (2019, methodology): “Our implementation uses the SEEDS [50] algorithm for superpixel computation.”
Street-view change detection with deconvolutional networks (2016, methodology): “We then extract a set of SEEDS superpixels [6], with size proportional to the scaled image, and ﬁt a plane for the support points belonging to each superpixel via RANSAC.”
Image Segmentation by Cascaded Region Agglomeration (2013, methodology): “Broadly, methods that produce superpixels can be grouped into graph-based [18, 6, 14, 22], clustering of pixels such as SLIC [1] and MeanShift [4], and curve evolution such as Turbopixels [13] and SEEDS[21].”
Superpixel segmentation using Linear Spectral Clustering (2015, methodology): “proposed SEEDS in [2] by introducing an energy function that encourages color homogeneity and shape regularity.”
High-Resolution Shape Completion Using Deep Neural Networks for Global Structure and Local Geometry Inference (2017, neutral): “The method in [27] is adopted to create holes.”
Online Video SEEDS for Temporal Window Objectness (2013, methodology): “In terms of its still counterparts, it comes closest to the recently introduced SEEDS superpixels [15].”
CNN-Enhanced Graph Convolutional Network With Pixel- and Superpixel-Level Feature Fusion for Hyperspectral Image Classification (2020, methodology): “In this section, we test several state-of-the-art superpixel segmentation methods on the CEGCN, including the SLIC [45], the quick shift (QS) [52], the Felzenszwalb and Huttenlocher’s (FH) method [53], the linear spectral clustering (LSC) [54], and the SEEDS [55].”
Superpixels: An evaluation of the state-of-the-art (2016, methodology): “From the ranking in Table 3, we recommend 6 algorithms for use in practice, thereby covering a wide range of application scenarios: ETPS [74], SEEDS [71], ERS [38], CRS [69, 70], ERGC [59] and SLIC [65].”
Stereo Video Object Segmentation Using Stereoscopic Foreground Trajectories (2019, methodology): “[48] also recovered incomplete depth data using superpixels [47].”
Beyond pixels: A comprehensive survey from bottom-up to semantic image segmentation and cosegmentation (2015, methodology): “Van den Berge et al. [68] proposed the fastest superpixel method-SEED, which can run at 30Hz.”
BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (2020, background): “…the threshold selection (Otsu, 1979), the region growing (Vin-cent and Soille, 1991), the super-pixel (Ren and Ma-lik, 2003; Achanta et al., 2012; Van den Bergh et al., 2012) and the graph (Boykov and Jolly, 2001; Rother et al., 2004) algorithms adopt the hand-crafted features to solve this…”
Superpixels and Polygons Using Simple Non-iterative Clustering (2017, methodology): “Both the plots show that SNIC exhibits linear complexity in practice and is also the fastest method after SEEDS.”