Virtual sample generation for template-based shape matching

Abstract

This paper presents a method for improving the performance of matching systems that correlate using shape templates. The basic idea involves extending an existing set of training shapes with generated "virtual" shapes, in order to improve representational capability. Yet no a-priori feature correspondence is necessary among the original shapes in the training set. Instead, an integrated clustering and registration approach partitions the original shape samples into clusters of similar and registered shapes; in each cluster a separate feature space is embedded. This allows for each cluster the derivation of standard compact parameterizations. This paper demonstrates that sampling these low-order spaces can produce an extended training set which facilitates a superior matching performance, as measured by a ROC curve. In the experiments, we consider a realistic application involving thousands of pedestrian shapes and perform correlation matching based on distance transforms.

A mixed generative-discriminative framework for pedestrian classification

“…Extensions to cover non-rigid shape deformations involved using a single global shape model in combination with nonlinear PCA techniques [25,26] or a probabilistic model [4], as well as the use of multiple local-linear shape models [7,11]. Other work considered joint linear shape and texture models [3,6,15] to capture underlying dependencies in a principled way.…”

“…Shape Model Texture Model Sample Plausibility Cootes et al [3] Fan et al [6] Jones et al [15] global linear (PCA) global linear (PCA) limit on deviation from mean Jones et al [14] multi-layer global linear (weighted PCA) multi-layer global linear (weighted PCA) limit on deviation from mean Gavrila et al [7] Heap et al [11] pose-specific linear (PCA) -limit on deviation from mean Romdhani et al [25] global non-linear (Kernel PCA) -limit on deviation from mean Sozou et al [26] global non-linear (polynomial spatial rearrangement of object parts [14]. See Table 1.…”

A Bayesian Framework for Multi-cue 3D Object Tracking

“…Only if this distance is lower than a user defined threshold, the shapes fall into the same cluster and the registration is assumed valid. For details, the reader is referred to [8].…”

“…Integrated registration and clustering: At first an integrated registration and clustering approach [8] is performed. The idea of integration is motivated by the fact, that general automatic registration methods are not able to find the physically correct point correspondences, if the variance in object appearance is too high.…”

Object Instance Sharing by Enhanced Bounding Box Correspondence

“…The simplest way of reusing instances is by perturbing existing instances [8,12,16,22] e.g., creating shifted, rotated, or mirrored versions. Most related is the recent work on instancesharing [10,18,19,28], with the key difference that our focus is on reusing instances within the same category and dataset.…”