Local Features and Kernels for Classification of Texture and Object Categories: A Comprehensive Study

By Jianguo Zhang, Marcin Marszalek, Svetlana Lazebnik, Cordelia Schmid. 2006

2,191 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Statistical Learning Theory

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

In Statistical Learning Theory

Citations in the corpus

CrossTransformers: spatially-aware few-shot transfer (2020, background): “Discriminative parts [5, 10, 22, 34, 42, 83] and visual words [74, 100] have a rich history, and have found applications in deformable-parts models [26], and classification [70, 100].”
On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation (2015, methodology): “Example case: χ(2) kernel The χ(2) kernel function has been applied with good performance in various image classification tasks [41, 54, 55].”
Bag of visual words and fusion methods for action recognition: Comprehensive study and good practice (2014, neutral): “For normalization methods, typical choices include l1normalization [64], l2-normalization [50], power normalization [35], and intra normalization [2].”
Approximating CNNs with Bag-of-local-Features models works surprisingly well on ImageNet (2019, methodology): “Many successful image classification models have been based on this pipeline (Csurka et al., 2004; Jurie & Triggs, 2005; Zhang et al., 2007; Lazebnik et al., 2006), see O’Hara & Draper (2011) for an up-to-date overview.”
Efficient Feature Extraction, Encoding, and Classification for Action Recognition (2014, methodology): “To recognize actions we use SVM with multi-channel exponential χ(2)-kernel [41] together with histogram-based action representations.”
Evaluation of Local Spatio-temporal Features for Action Recognition (2009, neutral): “A is the mean value of distances between all training samples [28].”
Multiple instance classification: Review, taxonomy and comparative study (2013, methodology): “A similar thing happens with the EMD and Gartner et al. [14] methods.”
Continuous Hyper-parameter Learning for Support Vector Machines (2015, methodology): “Moreover, for this task we use exponential χ2 kernels because they show naturally better performance on histogram data compared to RBF kernels [24].”
Learning The Discriminative Power-Invariance Trade-Off (2007, methodology): “Shape distances between two images are calculated as the χ 2 statistic between the normalised frequency histograms of densely sampled, vector quantised SIFT descriptors [29] of the two images.”
Picking the best DAISY (2009, background): “In recognition it is often combined with vector quantization to create “visual words” for searching large image databases and for object class recognition [14, 15, 24, 7, 4].”
Object Detection Using Strongly-Supervised Deformable Part Models (2012, neutral): “Bag-of-Features (BOF) models have been highly successful for object retrieval [18] and image classification [19].”
Affordance Prediction via Learned Object Attributes (2011, background): “The multi-channel 2 kernel has been shown to work well for the classification of image data, where multiple, unrelated, features are extracted and combined [31, 32].”
Efficient Classification for Additive Kernel SVMs (2013, background): “Index Terms—Image classification, support vector machines, efficient classifiers, additive kernels Ç”
Feature sampling and partitioning for visual vocabulary generation on large action classification datasets (2014, background): “where Hi and Hj are the histograms associated by videos i and j, K is the number of cluster centres, and A is the mean value of all distances between training histograms [29,18,13].”
Egocentric recognition of handled objects: Benchmark and analysis (2009, background): “Object recognition is a central problem in vision and has seen a huge amount of progress in the recent years [25], with highly influential works including [17, 3, 4, 9, 33].”
Local Higher-Order Statistics (LHS) for Texture Categorization and Facial Analysis (2012, methodology): “[18, 19] generate an image specific texton representation from rotation and scale invariant descriptors and compare them using Earth Movers distance, whereas [6,7,17,20] use a dictionary learned over the complete dataset to represent each image as histogram over this dictionary.”