Decoupled marginal distribution of gradient magnitude and laplacian of gaussian for texture classification

Wufeng Xue, Xuanqin Mou, Lei Zhang

Research output: Chapter in book / Conference proceedingConference article published in proceeding or bookAcademic researchpeer-review


We propose a novel descriptor for classification of texture images based on two isotropic low level features: the gradient magnitude (GM) and the Laplacian of Gaussian (LOG). The local descriptor is devised as the concatenation of the marginal distributions and a decoupled marginal distributions of the two features in local patch. The isotropic low level features and the computation of the two distributions ensure the rotation invariance and its robustness. To make the descriptors contrast invariant, within each image and across difference images of the same class, L2-normalization and Weber normalization are implied to the two features. After examined on three benchmark datasets, the proposed descriptor is showed to be more effective than other filter bank based features. Besides, the proposed descriptor can achieve very good performance even with small patch.
Original languageEnglish
Title of host publicationComputer Vision CCF Chinese Conference, CCCV 2015, Proceedings
PublisherSpringer Verlag
Number of pages11
ISBN (Print)9783662485576
Publication statusPublished - 1 Jan 2015
Event1st Chinese Conference on Computer Vision, CCCV 2015 - Xian, China
Duration: 18 Sept 201520 Sept 2015

Publication series

NameCommunications in Computer and Information Science
ISSN (Print)1865-0929


Conference1st Chinese Conference on Computer Vision, CCCV 2015


  • Decoupled marginal distributions
  • Gradient magnitude
  • Laplacian of gaussian
  • Texture classification

ASJC Scopus subject areas

  • Computer Science(all)
  • Mathematics(all)


Dive into the research topics of 'Decoupled marginal distribution of gradient magnitude and laplacian of gaussian for texture classification'. Together they form a unique fingerprint.

Cite this