Discriminative Random Fields

By Sanjiv Kumar, Martial Hebert. International Journal of Computer Vision, 2006

445 citations recorded by the source corpus.

In this research we address the problem of classification and labeling of regions given a single static natural image. Natural images exhibit strong spatial dependencies, and modeling these dependencies in a principled manner is crucial to achieve good classification accuracy. In this work, we present Discriminative Random Fields (DRFs) to model spatial interactions in images in a discriminative framework based on the concept of Conditional Random Fields proposed by lafferty et al.(2001). The DRFs classify image regions by incorporating neighborhood spatial interactions in the labels as well as the observed data. The DRF framework offers several advantages over the conventional Markov Random Field (MRF) framework. First, the DRFs allow to relax the strong assumption of conditional independence of the observed data generally used in the MRF framework for tractability. This assumption is too restrictive for a large number of applications in computer vision. Second, the DRFs derive their classification power by exploiting the probabilistic discriminative models instead of the generative models used for modeling observations in the MRF framework. Third, the interaction in labels in DRFs is based on the idea of pairwise discrimination of the observed data making it data-adaptive instead of being fixed a priori as in MRFs. Finally, all the parameters in the DRF model are estimated simultaneously from the training data unlike the MRF framework where the likelihood parameters are usually learned separately from the field parameters. We present preliminary experiments with man-made structure detection and binary image restoration tasks, and compare the DRF results with the MRF results.

Subfields: Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/79882a6e8da81f88a3705c1d276cdd8676f9cbe0

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Learning a Dictionary of Shape Epitomes with Applications to Image Labeling (2013, background): “Most modern image labeling systems are based on Conditional Random Fields (CRFs) [18, 20] for integrating local cues with neighborhood constraints.”
A coarse-to-fine model for airport detection from remote sensing images using target-oriented visual saliency and CRF (2015, background): “Conditional random field (CRF) has shown the flexible capacity to incorporate spatial contextual information in many computer vision problems, such as image semantic segmentation [25,26] and object detection [27].”
Top-down visual saliency via joint CRF and dictionary learning (2012, methodology): “Instead of using the pseudo-likelihood method [14], we take a discriminative training approach by converting the likelihood maximization into an inequality constrained optimization problem [11, 25].”
A Hybrid Approach to Detect and Localize Texts in Natural Scene Images (2011, methodology): “Any two components whose spatial relationship obeys this rule are linked as neighbors in the graph.”
Multi-scale context for scene labeling via flexible segmentation graph (2016, background): “DRF learns pairwise compatibilities including image information between labels of different nodes.”
A Discriminatively Trained Fully Connected Conditional Random Field Model for Blood Vessel Segmentation in Fundus Images (2017, methodology): “In the original definition of CRFs, images are mapped to graphs, where each pixel represents a node, and every node is connected with an edge to their neighbors according to a certain connectivity rule [29], [36], [39].”
Contextual classification with functional Max-Margin Markov Networks (2009, methodology): “CRFs are a popular tool in computer vision [13] for propagating contextual information among neighboring sites.”
Fields of Experts (2009, methodology): “For these applications, it can be beneficial to train MRF models discriminatively (Ning et al., 2005; Kumar and Hebert, 2006).”
Recent Advances on Spectral–Spatial Hyperspectral Image Classification: An Overview and New Guidelines (2018, background): “Besides, DRFs, having drawn broad attention in recent years [95], offer several advantages, namely: 1) the relaxation of conditional independence of the observed data; 2) the exploitation of probabilistic discriminative models instead of the generative MRFs; and 3) the simultaneous estimation of all”
The potential of automatic methods of classification to identify leaf diseases from multispectral images (2011, methodology): “This concept has been extended and made applicable to images by Kumar and Hebert (2006).”
Shape and Appearance Context Modeling (2007, background): “(2)Different flavors of HOG’s have proven to be successful in several settings [12, 2, 10].”
A conditional random field-based model for joint sequence segmentation and classification (2013, background): “During the last years, we have witnessed an explosion of interest in CRFs, as they have managed to achieve superb prediction performance in a variety of scenarios, thus being one of the most successful approaches to the structured output prediction problem, with successful applications including tex”
Robust Nucleus/Cell Detection and Segmentation in Digital Pathology and Microscopy Images: A Comprehensive Review (2016, background): “The CRF globally conditioned on X can be represented as [286]”
Learning Graphical Model Parameters with Approximate Marginal Inference (2013, methodology): “The pseudolikelihood has been used in many applications [30], [31].”
An Introduction to Conditional Random Fields (2010, methodology): “As an example application, we may wish to classify areas of a given input image according to whether they are foreground or background, they are a manmade structure or not [61, 62], or whether they are sky, water, vegetation, etc.”
Contextual classification of lidar data and building object detection in urban areas (2014, methodology): “In many CRF-based applications, the pairwise potentials are based on the probability of two neighbouring node labels yi and yj being identical given lijðxÞ (Kumar and Hebert, 2006): wijðx; yi; yjÞ / pðyi 1⁄4 yjjlijðxÞÞ: ð5Þ”