Multi-task low-rank affinity pursuit for image segmentation

Abstract

This paper investigates how to boost region-based im age segmentation by pursuing a new solution to fuse multi ple types of image features . A collaborative image segmen tation framework, called multi-task low-rank affinity pur suit, is presented for such a purpose . Given an image de scribed with multiple types of features, we aim at inferring a unified affinity matrix that implicitly encodes the segmen tation of the image . This is achieved by seeking the sparsity consistent low-rank affinities from the joint decompositions of multiple feature matrices into pairs of sparse and low rank matrices, the latter of which is expressed as the pro duction of the image feature matrix and its corresponding image affinity matrix . The inference process is formulated as a constrained nuclear norm and t'2,1-norm minimization problem, which is convex and can be solved efficiently with the Augmented Lagrange Multiplier method . Compared to previous methods, which are usually based on a single type of features, the proposed method seamlessly integrates mul tiple types of features to jointly produce the affinity matrix within a single inference step, and produces more accu rate and reliable segmentation results . Experiments on the MSRC dataset and Berkeley segmentation dataset well val idate the superiority of using multiple features over single feature and also the superiority of our method over conven tional methods for feature fusion . Moreover, our method is shown to be very competitive while comparing to other state-of-the-art methods .

Graph-based image segmentation using weighted color patch

“…In our experiments, we test the proposed method on all the 300 images, since the algorithm has no parameter to be trained. The number of segments k is set from [3,5,7,10,12,15,18,20,23,25,28,30,31,32,35,40]. The final results are evaluated according to 4 associated measurements, including: Probabilistic Rand Index (PRI) [21], Variation of Information (VoI) [17], Global Consis- tency Error (GCE) [16], and Boundary Displacement Error (BDE) [9].…”

“…Many kinds of features can be explored to compute the similarity of two nodes. For example, [15] used the averaged color to compute the affinity graph, and [5] fused color histogram, local binary patterns (LBP) and scale invariant feature transform (SIFT) with low-ranking.…”

Robust Object Co-detection

“…In particular, Liu et al [15] proposed the low-rank representation method which can be used to discover the underlying subspace structures by imposing the low-rank constraint on the representation coefficient matrix while using 2,1 -norm to remove outliers. We have recently seen several successful similar formulations of this toward applications including image segmentation [8] and saliency detection [24]. Our method is distinct in that we develop a low-rank coefficient matrix that is shared over multiple reconstructions derived from different features.…”

“…, K) and U are the Lagrange multipliers, and μ > 0 is a penalty parameter. For fast convergence speed, we use inexact ALM to solve (8) and resulting optimization procedure is found in Algorithm 1.…”

Efficient Image Segmentation Method Based on Sparse Subspace Clustering

“…Among them, a weighted sparse subspace clustering method is proposed for image segmentation by Li Tao and others [6], Zhang Wenjuan et al [7] proposed an image segmentation method with non convex low rank sparsity constraints method. Mode of (Localbinary pattern Cheng et [8] combined with a variety of image features, such as color histogram, color histograms, CH) and local binary LBP), based on the scale of the bow (bag of words) invariant feature transform SIFT-BoW (SIFT based onbag-of-words) etc. constitute a high dimensional feature, and the high dimensional feature through the low rank subspace clustering to achieve image segmentation, to a large number of natural image segmentation results show that this method can significantly improve the image segmentation results.…”