Shape matching and registration by data-driven EM

Abstract

In this paper, we present an efficient and robust algorithm for shape matching, registration, and detection. The task is to geometrically transform a source shape to fit a target shape. The measure of similarity is defined in terms of the amount of transformation required. The shapes are represented by sparse-point or continuous-contour representations depending on the form of the data. We formulate the problem as probabilistic inference using a generative model and the EM algorithm. But this algorithm has problems with initialization and computing the E-step. To address these problems, we define a data-driven technique (discriminative model) which makes use of shape features. This gives a hybrid algorithm which combines the generative and discriminative models. The resulting algorithm is very fast, due to the effectiveness of shape-features for solving correspondence requiring only a few iterations. We demonstrate the effectiveness of the algorithm by testing it on standard datasets, such as MPEG7, for shape matching and by applying it to a range of matching, registration, and foreground/background segmentation problems.

Layered Graph Matching with Composite Cluster Sampling

“…The overall hit rate and false alarm rate are reported in Fig. 19 with comparison to matching algorithms by Tu et al [31] and Liu et al [17]. The horse category has relatively lower detection rate due to articulations.…”

“…Recently, a similar method has also applied to unsupervised learning of object categories through matching parse trees across multiple images [30]. Some recent works on shape recognition, such as shape context [3] and shape matching [31], represent the graph structures implicitly for computational efficiency.…”

“…M ANY computer vision tasks can be posed as either a graph partitioning (or coloring) problem, such as image segmentation [1], [32] and scene labeling, or a graph matching (or correspondence) problem, such as wide baseline stereo [2], [20], [8], large motion [20], [26], [33], [35], object, and shape recognition [14], [3], [31], [37]. In this paper, we study a framework called layered graph matching for integrating graph partitioning and matching with their graphs edited.…”

Segmentation, Inference and Classification of Partially Overlapping Nanoparticles

“…Although the ECM-solution approach for shape classification and inference is not entirely new by itself [49], our proposed model and solution procedure can solve the morphology analysis problem for a large number of overlapping nanoparticles, evolving an equally large number of contours with guidance of multiple reference shapes. To our best knowledge, there is no other method that has such capability.…”

Parametric Estimation of Affine Deformations without Correspondences

“…As nding the exact maximum a posteriori is intractable, a linear programming relaxation technique is used. In [101], the registration problem is formulated as a probabilistic inference using a generative model and the expectation-maximization algorithm. The authors dene a data-driven technique which makes use of shape features.…”