Iterative figure-ground discrimination

Zhao, Liang; Davis, Larry S.

doi:10.1109/icpr.2004.1334006

Cited by 17 publications

(15 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Normalized Color Feature : 3D normalized rgs color 2 coordinates are commonly used as illumination insensitive features since the separation of chromaticity from brightness in the rgs space allows the use of a much wider kernel with the s variable to cope with the variability in brightness due to shading effects [23]. Color Rank Feature: The features are encoded as the relative rank 3 of intensities of each color channel R, G and B for all sample pixels.…”

Section: Appearance Modelmentioning

confidence: 99%

Learning Pairwise Dissimilarity Profiles for Appearance Recognition in Visual Surveillance

Lin

Davis

2008

Advances in Visual Computing

View full text Add to dashboard Cite

Abstract. Training discriminative classifiers for a large number of classes is a challenging problem due to increased ambiguities between classes. In order to better handle the ambiguities and to improve the scalability of classifiers to larger number of categories, we learn pairwise dissimilarity profiles (functions of spatial location) between categories and adapt them into nearest neighbor classification. We introduce a dissimilarity distance measure and linearly or nonlinearly combine it with direct distances. We illustrate and demonstrate the approach mainly in the context of appearance-based person recognition.

show abstract

Section: Appearance Modelmentioning

confidence: 99%

Learning Pairwise Dissimilarity Profiles for Appearance Recognition in Visual Surveillance

Lin

Davis

2008

Advances in Visual Computing

View full text Add to dashboard Cite

show abstract

“…, using a Sampling-Expectation (SE) algorithm proposed in [18]. We simultaneously and automatically compute the bandwidths (H i 's), and thresholds (τ i 's) corresponding to each layer (L i ), that are required for the ML-based pixel assignment (refer to Section 3).…”

Section: Trainingmentioning

confidence: 99%

“…This is done by a refinement step that uses the Sampling-Expectation (SE) technique proposed in [18]. There are 3 main steps in this refining process:…”

Section: Refinement Stepmentioning

confidence: 99%

“…For details about the concept of Kernel Density Estimation the reader is referred to [11]. Given a pixel y belonging to the image, we estimate the probabilities P Lc (y) and P bg (y) by first computing the following parameters (as in [18]): where, K is the kernel or smoothing function (we use a Gaussian kernel), d is the dimension of the feature-space (3 in our case, 5 if we incorporate optical flow) and h j 's are the kernel bandwidths, which we estimate using h j ≈ 1.06σ j m 1 5 [9], whereσ j is the standard deviation estimated over the sample S, in dimension j. The pixel probabilities are then computed as:…”

Section: Refinement Stepmentioning

confidence: 99%

See 1 more Smart Citation

Pixel Layering and Layer Propagation for Video Modelling and Foreground Detection

Patwardhan

Sapiro

Morellas

2005

View full text Add to dashboard Cite

show abstract

“…To avoid including background pixels in the person region, we do not use all pixels in the person's bounding box. Instead, we use a figure-ground expectation sampling (ES) technique [28] to segment the person from the background region ( Figure 5), and calculate the most discriminative features between the person and background. To encode spatial information, we subdivide the person region into a rectangular grid, and obtain the color distribution of each subregion using the most discriminative color features.…”

Section: Initializationmentioning

confidence: 99%

Using vision, acoustics, and natural language for disambiguation

Fransen

Morariu

Martinson

et al. 2007

Proceedings of the ACM/IEEE International Conference on Human-Robot Interaction

View full text Add to dashboard Cite

Creating a human-robot interface is a daunting experience. Capabilities and functionalities of the interface are dependent on the robustness of many different sensor and input modalities. For example, object recognition poses problems for state-of-the-art vision systems. Speech recognition in noisy environments remains problematic for acoustic systems. Natural language understanding and dialog are often limited to specific domains and baffled by ambiguous or novel utterances. Plans based on domain-specific tasks limit the applicability of dialog managers. The types of sensors used limit spatial knowledge and understanding, and constrain cognitive issues, such as perspective-taking.In this research, we are integrating several modalities, such as vision, audition, and natural language understanding to leverage the existing strengths of each modality and overcome individual weaknesses. We are using visual, acoustic, and linguistic inputs in various combinations to solve such problems as the disambiguation of referents (objects in the environment), localization of human speakers, and determination of the source of utterances and appropriateness of responses when humans and robots interact. For this research, we limit our consideration to the interaction of two humans and one robot in a retrieval scenario. This paper will describe the system and integration of the various modules prior to future testing.

show abstract

Iterative figure-ground discrimination

Cited by 17 publications

References 6 publications

Learning Pairwise Dissimilarity Profiles for Appearance Recognition in Visual Surveillance

Learning Pairwise Dissimilarity Profiles for Appearance Recognition in Visual Surveillance

Pixel Layering and Layer Propagation for Video Modelling and Foreground Detection

Using vision, acoustics, and natural language for disambiguation

Contact Info

Product

Resources

About