Creating efficient codebooks for visual recognition

By Frédéric Jurie, Bill Triggs. Tenth IEEE International Conference on Computer Vision (ICCV'05) Volume 1, 2005

853 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/8d32093cd04d6beffb6d757f58b5ac950543ff7d

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 Media Technology

Citations in the corpus

Fisher Kernels on Visual Vocabularies for Image Categorization (2007, methodology): “It has been observed that, even on databases containing a restricted number of categories (< 10), the best performance is generally obtained with large vocabularies containing from several hundreds to several thousands of visual words [2, 7, 11, 14, 15, 18].”
UvA-DARE ( Digital Academic Repository ) The visual extent of an object : suppose we know the object locations (2011, methodology): “In the final step, the quantity of visual words was found to be more important than the quality of the location of the visual words (Jurie and Triggs 2005; Nowak et al. 2006).”
Object Classification of Aerial Images With Bag-of-Visual Words (2010, background): “Owing to its simplicity, k-means [13], [14] is exploited in this letter for visual-word encoding [15].”
Evaluation of Local Spatio-temporal Features for Action Recognition (2009, neutral): “Moreover, we compare with dense features obtained by regular sampling of local space-time patches, as recently excellent results were obtained by dense sampling in the context of object recognition [7, 11].”
Evaluating Color Descriptors for Object and Scene Recognition (2010, neutral): “The bag-of-words model is also known as “textons” [30], “object parts” [31], and “codebooks” [32], [33].”
Visual Word Ambiguity (2010, background): “The most frequent features, as noted by Jurie and Triggs [12] and others [3], [23], are not necessarily the most discriminative.”
Unsupervised object discovery via self-organisation (2012, methodology): “To overcome these problems, Jurie and Triggs (2005) added a radius term to k-means in order to enforce the cluster centres to cover the input space better, and they also proposed a method to sort the codebook bins to make them meaningful.”
Sampling Strategies for Bag-of-Features Image Classification (2006, neutral): “…F¨orstner, Harris-aﬃne, etc.) as samplers [4, 1, 2, 5, 6, 7, 8, 9, 10, 11], but although such detectors have proven their value in matching applications, they were not designed to ﬁnd the most informative patches for image classiﬁcation and there is some evidence that they do not do so [12, 13].”
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.”
50 Years of object recognition: Directions forward (2013, background): “They represent an attempt to compensate for the simplicity of localfeature-based representations, by using them to form dictionaries of more complex and compact object representations [195, 196, 179, 177, 176, 197, 178, 180, 198, 199, 100, 174, 200, 201, 202, 203].”
A Comparative Study of Sampling Analysis in the Scene Classification of Optical High-Spatial Resolution Remote Sensing Imagery (2015, background): “However, it has been proven that such keypoint detectors cannot improve the classification performance [35,36], because they were not designed for classification problems, but rather for image matching applications.”
The 2005 PASCAL Visual Object Classes Challenge (2005, neutral): “Second, the dense feature set is extremely unbalanced as was shown in [24].”
Locality-constrained Linear Coding for image classification (2010, methodology): “Of the many extensions of the BoF method, including the generative part models [7, 3, 2], geometric correspondence search [1, 14] and discriminative codebook learning [13, 17, 23], the most successful results were reported Feature vector [ ]”
Image classification by non-negative sparse coding, low-rank and sparse decomposition (2011, background): “Some tried to learn discriminative visual codebooks for image classification [12, 13].”
Fast Discriminative Visual Codebooks using Randomized Clustering Forests (2006, background): “K-means clustering is currently the most common [5, 22] but mean-shift [9] and hierarchical k-means [17] clusterers have some advantages.”
Incremental learning of nonparametric Bayesian mixture models (2008, background): “to build dictionaries of visual words in order to train recognition models for textures, objects, and scenes [13, 20, 5, 7, 10].”