Abstract:Many computer vision tasks can be posed as either a graph part itioning (segmentation) problem or a graph matching (correspondence) problem. In this paper, we present an integrated solution for simultaneous graph partition and matching with graph editing. Our object ive is to find and match an unknown number of common graph structures (objects) in two images – a problem a rising in recent object recognition research, such as object categorization and unsupervised learning. U like previous segmentation and correspondence problems motivated by large motion or wide baseline stereo, the scenes and objects in the two images may have quite different appearance but share similar graphic s tru tures. Given two images, we first extract their primal sketches [11] as attributed graphs which repre sent both textures (by key points) and structures (by line segments and primitives). The points, line segment s, and primitives are vertices of the graphs. Both graphs are partitioned into K + 1 layers of subgraphs so that K pairs of subgraphs are matched and the remaining layer contains unmatched background. Eac h p ir of matched subgraphs represent a common object under geometric deformations and structural editing with graph operators. We design a data-driven Markov Chain Monte Carlo (DDMCMC) algorithm to explore the joint partition and matching space effectively. It has two iterative components: (i) A Sw endsen-Wang Cut component in the partition space that can flip, in each step, a chunk (connected componen t) of the graph where the vertices are often strongly coupled due to proximity and similar appeara nce; (ii) A Gibbs sampling step for matching some connected components to the other graph. To prune the ma tching space, we grow a number of vertices to form some connected components so that each conn e ted component has a small number of candidate matches. We apply the algorithm to a number of appl ications with comparison to the state of the art methods: (i) multi-object wide-baseline matching (und erlying both rigid and non-rigid motions) with occlusions , (ii) automatic detection and recognition of co mmon objects from templates, and (iii) human articulate motion analysis between images from a video sequ ence. Index Terms Graph Matching, Graph Partitioning, DDMCMC, Motion Analys is, Swendsen-Wang Cut. This manuscript is submitted to IEEE Trans. on PAMI. A short version was published in CVPR2007 [14]. email contacts: llin@lotushill.org, xbliu@lotushill.orgxbliu@lotushill.org, sczhu@stat.ucla.edu

Parsing Images into Regions, Curves, and Curve Groups

Computing three-dimensional scene from a single image by bottom-up/top-down bayesian inference

Interactive Color Image Segmentation by Region Growing Combined with Image Enhancement Based on Bezier Model

Region competition: unifying snakes, region growing, energy/Bayes/MDL for multi-band image segmentation

Bayesian Inference for Layer Representation with Mixed Markov Random Field

Image segmentation by data-driven markov chain monte carlo

Generalizing Swendsen–Wang for Image Analysis

Modeling Complex Motion: Photometric, Geometric, Dynamic, and Topological Aspects

Region Competition: Unifying Snakes, Region Growing, and Bayes/MDL for Multiband Image Segmentation

Image Parsing: Unifying Segmentation, Detection, and Recognition.

On the relationship between image and motion segmentation

Image segmentation by MAP-ML estimations.

Bayesian Reconstruction of 3d Shapes and Scenes from A Single Image

A Computational Model of Topological and Geometric Recovery for Visual Curve Completion.

Mpetition: Unifying Snakes, Egion Growin and Bayes/MDL for Image Segmentation

Effective Bayesian Inference by Data-Driven Markov Chain Monte Carlo for Object Recognition and Image Segmentation

Generalizing Swendsen-Wang to Sampling Arbitrary Posterior Probabilities

Integrating Graph Partition and Matching with Editing

How the brain learns to parse images using an attentional, incremental grouping process

Cluster sampling and its applications to segmentation, stereo and motion

Image Representation by Active Curves