Abstract:We investigate a fundamental problem in computer vision: unsupervised image segmentation. During the last decade, the Normalized Cuts has become very popular for image segmentation. NCuts guarantees a globally optimal solution in the continuous solution space, however, how to automatically select the number of segments for a given image is left as an open problem. Recently, the lossy minimum description length (LMDL) criterion has been proposed for segmentation of images. This criterion can adaptively determine the number of segments, however, as the optimization is combinatorial, only a suboptimal solution can be achieved by a greedy algorithm. The complementarity of both criteria motivates us to combine NCuts and LMDL into a unified fashion, to achieve a better segmentation: given the NCuts segmentations under different numbers of segments, we choose the optimal segmentation to be the one that; minimizes the overall coding length, subject to a given distortion. We then develop a new way to use the coding length decrement as the similarity measure for NCuts, so that our algorithm is able to seek both the optimal NCuts solution under fixed number of segments, and the optimal LMDL solution among different numbers of segments. Extensive experiments demonstrate the effectiveness of our algorithm.

Fast Normalized Cut with Linear Constraints

Normalized cut segmentation algorithm combined with fractional differential.

A Novel Normalized-Cut Solver with Nearest Neighbor Hierarchical Initialization

Spectral Graph Cut from a Filtering Point of View

Multi-Class Constrained Normalized Cut with Hard, Soft, Unary and Pairwise Priors and Its Applications to Object Segmentation

Convergence of Projected Subgradient Method with Sparse or Low-Rank Constraints

Fast Algorithm for Constrained Linear Inverse Problems

A New Relaxation Approach to Normalized Hypergraph Cut

Toward the Optimization of Normalized Graph Laplacian

Spectral Normalized-Cut Graph Partitioning with Fairness Constraints

Efficiently Solving the Piecewise Constant Mumford-Shah Model using Graph Cuts

Fast Low-Rank Subspace Segmentation

Scalable Constrained Spectral Clustering

Segmentation Via Ncuts And Lossy Minimum Description Length: A Unified Approach

Graph matching based on fast normalized cut and multiplicative update mapping

A Feasible Method for Optimization with Orthogonality Constraints

Spectrally Constrained Optimization

Explored Normalized Cut With Random Walk Refining Term for Image Segmentation

Fast gradient clustering

Norm-induced Cuts: Optimization with Lipschitzian Black-box Functions

A Sparse Graph Formulation for Efficient Spectral Image Segmentation