Texture classification: are filter banks necessary?

By Manik Varma, Andrew Zisserman. 2003

538 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition

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

Describing Textures in the Wild (2013, methodology): “We also evaluate as local descriptors the Leung and Malik (LM) [21] (48-D) and MR8 (8-D) [14, 39] filter banks, the 3 × 3 and 7 × 7 raw image patches of [38], and the local binary patterns (LBP) of [25].”
Invariant Scattering Convolution Networks (2012, methodology): “The lowest published classification errors obtained on this dataset are 2 percent for Markov random fields [35], 1.”
Character Recognition in Natural Images (2009, methodology): “For Kannada, we learn 3 words per class, resulting in a vocabulary of 1971 words.”
Content based image retrieval using motif cooccurrence matrix (2004, methodology): “The author presented a novel approach [38] for capturing images having similar textures using the joint distribution of intensity values over a small image patch (as small as 3 £ 3 pixel square).”
Viewpoint Invariant Texture Description Using Fractal Analysis (2009, methodology): “The results for the methods (Lazebnik et al. 2005) and (Varma and Zisserman 2002) are from the paper Lazebnik et al. (2005), and the results for Varma and Garg’s method are from their paper (Varma and Garg 2007).”
Mean shift based clustering in high dimensions: a texture classification example (2003, background): “Since textons capture local spatial configurations, we believe that combining the mode textons with the representation proposed in [19] can offer more insight into why the texton approach is superior to previous techniques.”
Deep Filter Banks for Texture Recognition, Description, and Segmentation (2015, background): “As textures are imaged under variable truncation, viewpoint, and illumination, these datasets have stimulated the creation of texture representations that are invariant to viewpoint and illumination changes (Varma and Zisserman 2005; Ojala et al. 2002; Varma and Zisserman 2003; Leung and Malik 2001)”
Classifying materials in the real world (2010, methodology): “3) that using SVMs leads to a gain in performance compared to the method proposed by Varma and Zisserman [5,9].”
Scene Classification Using a Hybrid Generative/Discriminative Approach (2008, methodology): “Previously, both sparse [7], [17], [28] and dense descriptors, for example, [8], [19], [31] have been used.”
Steel defect classification with Max-Pooling Convolutional Neural Networks (2012, background): “However, in a broader context the problem can be viewed as defect detection in textured material which has received considerable attention in computer vision [7], [8], [9].”
A Bayesian hierarchical model for learning natural scene categories (2005, methodology): “We could loosely think of a texture as one particular intermediate representation of a complex scene.”
A Completed Modeling of Local Binary Pattern Operator for Texture Classification (2010, unknown): “Recently, Varma and Zisserman [6] proposed to learn a rotation invariant texton dictionary from a training set, and then classify the texture image based on its texton distribution.”
Deep convolutional filter banks for texture recognition and segmentation (2014, methodology): “Typical combinations of the ﬁlter responses, identiﬁed by vector quantisation, were used as the computational basis of the “textons” proposed by Julesz [22].”
A Training-free Classification Framework for Textures, Writers, and Materials (2012, methodology): “In the literature, the local texture descriptors are extracted over a sparse set of interest points [14], over a denser grid [13], or, more commonly, in every point in the image [20, 23, 31, 32].”
Discriminative learned dictionaries for local image analysis (2008, background): “, by designing local texture models [15, 37], which have proven to be discriminative enough to be used in image segmentation [18, 33].”
Invariant Scattering Convolution Networks (2013, methodology): “The lowest published classification errors obtained on this dataset are 2 percent for Markov random fields [35], 1.”
Local Features and Kernels for Classification of Texture and Object Categories: A Comprehensive Study (2006, methodology): “For the UIUCTex database we can observe that both our method and Lazebnik’s method work much better than Hayman’s method and VZ-joint, while Hayman’s method works better than VZ-joint.”