Content-sensitive superpixel segmentation via self-organization-map neural network

Wang, Murong; Liu, Xiabi; Soomro, Nouman Qadeer; Han, Guanghui; Liu, Weihua

doi:10.1016/j.jvcir.2019.102572

Cited by 8 publications

(5 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…According to Lin et al [18], they can be considered low-level image information subdivisions. As stated by [17], [19], [20], the use of superpixels brings several advantages: it reduces the size of the classification problem, since one cluster represents many pixels; it allows for a richer feature representation; and it produces homogeneous regions with additional semantics despite being an oversegmentation.…”

Section: Open-set Segmentation Methodsmentioning

confidence: 99%

“…Nevertheless, non-neighboring superpixels may have similar values and shapes. As examples of proposed techniques in the last two decades: Felzenszwalb [17]; Quickshift [36]; TurboPixels [37]; ERS [38]; SLIC [39]; GSM [40]; Eikonal-based [41]; SEEDS [42]; LSC [43]; Waterpixels [44]; BASS [45]; SAS [46]; SH+FDAG [20]; content-based [47]; SPFCM [48].…”

Section: Superpixel Segmentationmentioning

confidence: 99%

See 1 more Smart Citation

FuSC: Fusing Superpixels for Improved Semantic Consistency

Nunes,

Pereira,

Oliveira

et al. 2024

IEEE Access

View full text Add to dashboard Cite

Open-set segmentation has caught the community's attention only in the last few years, and it is a growing and active research area with many challenges ahead. To better identify open-set pixels, we address two known issues by improving data representation and ensuring semantic consistency in open-set predictions. First, we present a method called Open Gaussian Mixture of Models (OpenGMM) that allows for multimodal statistical distributions in known class pixels using a Gaussian Mixture of Models instead of unimodal approaches, like Principal Component Analysis. The second approach improved semantic consistency by applying a post-processing technique that uses superpixels to enforce homogeneous regions to have similar predictions, rectifying erroneously classified pixels within these regions and providing better delineation of object borders. We also developed a novel superpixel method called Fusing Superpixels for Improved Semantic Consistency (FuSC) that produced more homogeneous superpixels and enhanced, even more, the open-set segmentation prediction. We applied the proposed approaches to well-known remote sensing datasets with labeled ground truth for semantic segmentation tasks. The proposed methods improved the highest AUROC quantitative results for the ISPRS Vaihingen and Potsdam datasets. Using FuSC, we achieved novel open-set state-of-the-art results for both datasets, improving AUROC results from 0.850 to 0.880 (3.53%) for Vaihingen and 0.764 to 0.797 (4.32%) for Potsdam datasets. The official implementation is available at: https://github.com/iannunes/FuSC

show abstract

Section: Open-set Segmentation Methodsmentioning

confidence: 99%

Section: Superpixel Segmentationmentioning

confidence: 99%

FuSC: Fusing Superpixels for Improved Semantic Consistency

Nunes,

Pereira,

Oliveira

et al. 2024

IEEE Access

View full text Add to dashboard Cite

show abstract

“…SPSs are an active research area, and many different methods have been proposed to generate superpixels from an image. As examples of relevant methods proposed in the last 2 decades, we can cite: Felzenszwalb [18], Quickshift [38], TurboPixels [39], ERS [40], SLIC [41], GSM [42], Eikonalbased [43], SEEDS [44], LSC [45], Waterpixels [46], BASS [47], SAS [48], SH+FDAG [20], content-based [49] and SPFCM [50].…”

Section: B Open Set Semantic Segmentationmentioning

confidence: 99%

“…Superpixels are a oversegmentation of an image and played an important role in the process of traditional segmentation, and acoording to Lin et al [17] are a low-level image information subdivisions. As stated by [18]- [20] the use of superpixels: allows for a richer feature representation for each segment (textural, color, scatter, gradient-based, statistic, orientation); reduces the size of the classification problem since many pixels could be represented by each cluster; even being an oversegmentation of the image the superpixel clustering produces homogeneous regions that adds semantics to each superpixel. Superpixels are a homogeneous and contiguous group of pixels in an image extracting perceptually relevant regions [18].…”

Section: Introductionmentioning

confidence: 99%

FuSS: Fusing Superpixels for Improved Segmentation Consistency

Nunes¹,

Pereira²,

Oliveira³

et al. 2022

Preprint

View full text Add to dashboard Cite

In this work, we propose two different approaches to improve the semantic consistency of Open Set Semantic Segmentation. First, we propose a method called OpenGMM that extends the OpenPCS framework using a Gaussian Mixture of Models to model the distribution of pixels for each class in a multimodal manner. The second approach is a post-processing which uses superpixels to enforce highly homogeneous regions to behave equally, rectifying erroneous classified pixels within these regions, we also proposed a novel superpixel method called FuSS. All tests were performed on ISPRS Vaihingen and Potsdam datasets, and both methods were capable to improve quantitative and qualitative results for both datasets. Besides that, the post-process with FuSS achieved state-of-the-art results for both datasets. The official implementation is available at: https://github.com/iannunes/FuSS.

show abstract

“…SPSs are an active research area, and many distinct methods were proposed to generate superpixels from an image. As examples of well-known methods proposed in the last two decades, we can cite: Felzenszwalb (Felzenszwalb and Huttenlocher, 2004), Quickshift (Vedaldi and Soatto, 2008), TurboPixels (Levinshtein et al, 2009), ERS (Liu et al, 2011), SLIC (Achanta et al, 2012), GSM (Morerio et al, 2014), Eikonal-based (Buyssens et al, 2014), SEEDS (Bergh et al, 2012), LSC (Li and Chen, 2015), Waterpixels (Machairas et al, 2015), BASS (Rubio et al, 2016), SAS (Achanta et al, 2018), SH+FDAG (Wang et al, 2019), content-based (Zhang et al, 2020) and SPFCM (Elkhateeb et al, 2021). Among all possible choices of SPS algorithms to use in this work, we choose three algorithms that have fundamentally different strategies to generate the superpixels: SLIC (Achanta et al, 2012), Quickshift (Vedaldi and Soatto, 2008) and Felzenszwalb (Felzenszwalb and Huttenlocher, 2004).…”

Section: Superpixel Segmentationmentioning

confidence: 99%

Open-Set Semantic Segmentation for Remote Sensing Images

NUNES¹

View full text Add to dashboard Cite

show abstract

Content-sensitive superpixel segmentation via self-organization-map neural network

Cited by 8 publications

References 14 publications

FuSC: Fusing Superpixels for Improved Semantic Consistency

FuSC: Fusing Superpixels for Improved Semantic Consistency

FuSS: Fusing Superpixels for Improved Segmentation Consistency

Open-Set Semantic Segmentation for Remote Sensing Images

Contact Info

Product

Resources

About