Normalized Cuts for image segmentation
E1017405
computer vision technique
graph-based clustering method
image segmentation algorithm
spectral clustering method
Normalized Cuts for image segmentation is a graph-based computer vision technique that partitions an image into meaningful regions by optimizing a global criterion balancing inter-group dissimilarity and intra-group similarity.
Statements (51)
| Predicate | Object |
|---|---|
| instanceOf |
computer vision technique
ⓘ
graph-based clustering method ⓘ image segmentation algorithm ⓘ spectral clustering method ⓘ |
| advantage |
avoids bias toward small isolated regions
ⓘ
incorporates global image information ⓘ produces globally optimal partitions under relaxation ⓘ |
| alsoKnownAs |
Ncut
NERFINISHED
ⓘ
Normalized Cuts NERFINISHED ⓘ |
| appliedIn |
medical image analysis
ⓘ
object segmentation ⓘ scene segmentation ⓘ |
| assumes | image regions correspond to coherent groups in feature space ⓘ |
| balances |
between-group dissimilarity
ⓘ
within-group similarity ⓘ |
| basedOn |
graph theory
ⓘ
spectral graph theory ⓘ |
| coreConcept |
balancing inter-group dissimilarity and intra-group similarity
ⓘ
minimizing a normalized cut cost function ⓘ partitioning a graph into disjoint sets ⓘ |
| edgeWeightRepresents | similarity between pixels or regions ⓘ |
| field |
computer vision
ⓘ
image processing ⓘ pattern recognition ⓘ |
| graphMatrix |
affinity matrix
ⓘ
degree matrix ⓘ graph Laplacian ⓘ |
| influenced |
graph-based image segmentation methods
ⓘ
later spectral clustering algorithms ⓘ |
| introducedBy |
Jianbo Shi
NERFINISHED
ⓘ
Jitendra Malik NERFINISHED ⓘ |
| limitation |
computationally expensive for large images
ⓘ
requires construction of large affinity matrix ⓘ |
| objectiveFunction | normalized cut value ⓘ |
| objectiveMinimizes | sum of weights of edges cut between groups normalized by association within groups ⓘ |
| operatesOn |
image pixels
ⓘ
image regions ⓘ |
| publicationYear | 2000 ⓘ |
| publishedIn | IEEE Transactions on Pattern Analysis and Machine Intelligence NERFINISHED ⓘ |
| relatedTo |
graph partitioning
ⓘ
minimum cut ⓘ ratio cut ⓘ |
| representsImageAs | weighted undirected graph ⓘ |
| segmentationObtainedBy |
clustering in spectral embedding space
ⓘ
thresholding eigenvector components ⓘ |
| similarityCanBeBasedOn |
color
ⓘ
intensity ⓘ spatial proximity ⓘ texture ⓘ |
| solvedBy | generalized eigenvalue problem ⓘ |
| uses | eigenvectors of graph Laplacian ⓘ |
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.