Discovering Shape Classes using Tree Edit-Distance and Pairwise Clustering

Torsello, Andrea; Robles-Kelly, Antonio; Hancock, Edwin R.

doi:10.1007/s11263-006-8929-y

Cited by 43 publications

(22 citation statements)

References 48 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Image matching using graph image representations may be performed by: (a) exploiting spectral properties of the graphs' adjacency matrices [44,50,51]; (b) minimizing the graph edit-distance [12,47,62]; (c) finding a maximum clique of the association graph [41]; (d) using energy minimization or expectation-maximization of a statistical model [23,63]. All these formulations can be cast as a quadratic assignment problem, where a linear term in the objective function encodes node compatibility functions, and a quadratic term encodes edge compatibility functions.…”

Section: Object Discovery As Graph Matchingmentioning

confidence: 99%

From Region Based Image Representation to Object Discovery and Recognition

Ahuja

Todorovic

2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. This paper presents an overview of the work we have done over the last several years on object recognition in images from region-based image representation. The overview focuses on the following related problems: (1) discovery of a single 2D object category frequently occurring in a given image set; (2) learning a model of the discovered category in terms of its photometric, geometric, and structural properties; and (3) detection and segmentation of objects from the category in new images. Images in the given set are segmented, and then each image is represented by a region graph that captures hierarchy and neighbor relations among image regions. The region graphs are matched to extract the maximally matching subgraphs, which are interpreted as instances of the discovered category. A graph-union of the matching subgraphs is taken as a model of the category. Matching the category model to the region graph of a new image yields joint object detection and segmentation. The paper argues that using a hierarchy of image regions and their neighbor relations offers a number of advantages in solving (1)-(3), over the more commonly used point and edge features. Experimental results, also reviewed in this paper, support the above claims. Details of our methods as well of comparisons with other methods are omitted here, and can be found in the indicated references.

show abstract

Section: Object Discovery As Graph Matchingmentioning

confidence: 99%

From Region Based Image Representation to Object Discovery and Recognition

Ahuja

Todorovic

2010

Lecture Notes in Computer Science

View full text Add to dashboard Cite

show abstract

“…The lack of analytical methods in the domain of graphs led to the formulation of clustering problem as pairwise clustering [23] [17]. The process of pairwise clustering is somewhat different to the more familiar one of central clustering.…”

Section: Graph Based Clusteringmentioning

confidence: 99%

Clustering Using Class Specific Hyper Graphs

Xia

Hancock

2008

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Abstract. The aim in this paper is to develop a method for clustering together image views of the same object class. Local invariant feature methods, such as SIFT, have been proven effective for image clustering. However, they have made either relatively little use or too complex use of geometric constraints and are confounded when the detected features are superabundant. Here we make two contributions aimed at overcoming these problems. First, we rank the SIFT points (R-SIFT) using visual saliency. Second, we use the reduced set of R-SIFT features to construct a specific hyper graph (CSHG) model of holistic-structure. Based on the CSHG model, a two stage clustering method is proposed. In which, images are clustered according to the pairwise similarity of the graphs, which is a combination of the traditional similarity of local invariant feature vectors and the geometric similarity between two graphs. This method comprehensively utilizes both SIFT and geometric constraints, and hence combines both global and local information. Experiments reveal that the method gives excellent clustering performance.

show abstract

“…A number of approaches, including our previous work, use image contours as features [11,[13][14][15][16][17][18][19][20][21][22][23][24][25]. These methods argue that contours are in general richer descriptors, more discriminative, and more noise-tolerant than interest points.…”

Section: Introductionmentioning

confidence: 99%

Shape-Based Object Discovery in Images

Todorovic

Payet

2013

Shape Perception in Human and Computer Vision

View full text Add to dashboard Cite

Abstract. This paper presents an overview of our recent work on shape-based object recognition in images. The overview focuses on the following related problems: i) discovery of all distinct 2D object categories frequently occurring in an unlabeled set of images; ii) learning a model of the discovered categories; and iii) recognition and localization of objects from the discovered categories in new images. The paper argues that using image contours as basic features, and thus directly grounding object discovery and recognition on shape, offers a number of advantages in solving (i)-(iii) over more commonly used point features. Since shape is directly encoded by layouts of image contours, similar contour layouts across the images are expected to belong rather to object occurrences, than the background. The contour layouts are captured by a graph over all pairs of matching contours from different images. The graph's maximum a posteriori multicoloring assignment is taken to represent the shapes of discovered objects. Our empirical evaluation suggests that shape is more expressive and discriminative than photometric features for object discovery.

show abstract

Discovering Shape Classes using Tree Edit-Distance and Pairwise Clustering

Cited by 43 publications

References 48 publications

From Region Based Image Representation to Object Discovery and Recognition

From Region Based Image Representation to Object Discovery and Recognition

Clustering Using Class Specific Hyper Graphs

Shape-Based Object Discovery in Images

Contact Info

Product

Resources

About