View-based 3-D object recognition using shock graphs

Abstract

The shock graph is an emerging shape representation for object recognition, in which a 2-D silhouette is decomposed into a set of qualitative parts, captured in a directed acyclic graph. Although a number of approaches have been proposed for shock graph matching, these approaches do not address the equally important indexing problem. We extend our previous work in both shock graph matching and hierarchical structure indexing to propose the first unified framework for view-based 3-D object recognition using shock graphs. The heart of the framework is an improved spectral characterization of shock graph structure that not only drives a powerful indexing mechanism (to retrieve similar candidates from a large database), but also drives a matching algorithm that can accommodate noise and occlusion. We describe the components of our system and evaluate its performance using both unoccluded and occluded queries. The large set of recognition trials (over 25,000) from a large database (over 1400 views) represents one of the most ambitious shock graph-based recognition experiments conducted to date.

Object categorization using bone graphs

“…The contributions of the paper are twofold: (1) we introduce a novel framework for inexact graph matching, extending our body of previous work on matching node-attributed graphs [41,38,24,37] to accommodate node-and edge-attributed graphs, and expanding the range of domain-dependent topological constraints that can be enforced during matching; and (2) we apply the matching framework to a powerful new structured shape representation, the bone graph [23], yielding the first end-to-end object categorization system based on bone graphs. The improved stability of the node-and edge-attributed bone graph, combined with our matching algorithm's ability to exploit both its edge attributes and its domain-dependent topological constraints, yield an object categorization system which outperforms a state-of-theart shape categorization system based on shock graphs.…”

“…There is a large body of work in the area of inexact graph matching, but most of it is focused on graphs in which either the nodes or the edges are attributed. Our experience with the inexact graph matching algorithm proposed by Shokoufandeh et al [24,37,38,41,44] for DAGs with attributed nodes motivates our extension of this approach in order to incorporate edge information into the matching problem. We propose a generalization of this algorithm that expands the range of constraints that can be accounted for at matching time, leading to a general framework for representing domain knowledge about the relevance of structural (a) (b) (c) (d) Fig.…”

A survey of content based 3D shape retrieval methods

“…Using shock matching, Macrine et al [52] apply indexing using topological signature vectors to implement view based similarity matching more efficiently. Also, recently, view based similarity has been applied to retrieve 3D objects by Chen et al [17].…”

Object Categorization and the Need for Many-to-Many Matching

“…Moreover, the dimensionality of the resulting vectors is bounded by the maximum branching factor of the graph and not by the number of nodes in the graph. Not only do these structural signature vectors offer an effective mechanism for rapid indexing from large databases of graphs [7], but they offer a mechanism for matching graph abstractions [8,6,5]. If the vectors computed for two nodes of two directed acyclic graphs are similar, then their underlying subgraphs have similar structure, i.e., the collection of nodes forming the subgraph rooted at a node in one graph matches a collection of nodes forming the subgraph rooted at a node in a second graph, effectively yielding a many-to-many node (and structure) correspondence.…”