Segmentation using superpixels: A bipartite graph partitioning approach
read more
Citations
A survey of graph theoretical approaches to image segmentation
Lazy Random Walks for Superpixel Segmentation
Bipartite graph partitioning and data clustering
Real-Time Superpixel Segmentation by DBSCAN Clustering Algorithm
Locally Weighted Ensemble Clustering
References
Normalized cuts and image segmentation
Normalized cuts and image segmentation
Mean shift: a robust approach toward feature space analysis
On Spectral Clustering: Analysis and an algorithm
Related Papers (5)
Frequently Asked Questions (10)
Q2. What are the future works in "Segmentation using superpixels: a bipartite graph partitioning approach" ?
Future work should study the selection of superpixels more systematically and the incorporation of high-level cues. The following lemma will be used later to establish correspondence between the eigenvalues of ( 4 ) and ( 5 ). The following theorem states that an eigenvector of ( 4 ), if confined to Y, will be an eigenvector of ( 5 ). The converse of Theorem 2 is also true, namely, an eigenvector of ( 5 ) can be extended to be one of ( 4 ), as detailed in the following theorem.
Q3. How long does it take to generate a superpixel?
On average, SAS takes 6.44 seconds to segment an image of size 481×321, where 4.11 seconds are for generating superpixels, and only 0.65 seconds for the bipartite graph partitioning with Tcut.
Q4. What is the contribution of the authors?
Their another contribution is the development of a highly efficient spectral algorithm for bipartite graph partitioning, which is tailored to bipartite graph structures and provably much faster than conventional approaches, while providing novel insights on spectral clustering over bipartite graphs.
Q5. Why do the authors never encounter such cases in their experiments?
the authors never encounter such cases in their experiments probably because of the relatively strong connections between pixels and the superpixels containing them.
Q6. What is the simplest way to solve a partial SVD?
This partial SVD, done typically by an iterative process alternating between matrix-vector multiplications B̄v and B̄⊤u, is essentially equivalent to the Lanczos method on W̄ [13].
Q7. What is the simplest way to partition a graph?
The authors summarize their approach to bipartite graph partitioning in Algorithm 1, which the authors call Transfer Cuts (Tcut) since it transfers the eigen-operation from the original graph to a smaller one.
Q8. How do the authors achieve high performance on the Berkeley Segmentation Database?
The authors achieve highly competitive results on the Berkeley Segmentation Database with large performance gains ranging from 0.8146 to 0.8319 in PRI, and 12.21 to 11.29 in BDE, and very close to the best one in VoI and GCE (Section 4).
Q9. What is the simplest way to prove Lf = Df?
To prove Lf = γDf , it suffices to show (a) DXu − Bv = γDXu and (b) DY v − B⊤u = γDY v. (a) holds because u = 11−γD −1 X Bv is known.
Q10. What criteria are used to quantify the segmentation results?
To quantify the segmentation results, the authors follow common practice (e.g. [15, 9, 28]) to use the four criteria: 1) Probabilistic Rand Index (PRI) [27], measuring the likelihood of a pair of pixels being grouped consistently in two segmentations;