Open AccessLocal ZigZag Max histograms of pooling pattern for texture classification
Author(s) -
Roy Swalpa Kumar,
Dubey Shiv Ram,
Chaudhuri Bidyut B.
Publication year - 2019
Publication title -
electronics letters
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.375
H-Index - 146
ISSN - 1350-911X
DOI - 10.1049/el.2018.7631
Subject(s) - zigzag , histogram , pooling , artificial intelligence , pattern recognition (psychology) , texture (cosmology) , robustness (evolution) , pixel , mathematics , similarity (geometry) , image texture , computer vision , segmentation , image (mathematics) , computer science , image segmentation , geometry , biochemistry , chemistry , gene
The efficiency of any texture classification model confides on descriptor used for similarity matching. The formation of image descriptor is a challenging and important task in computer vision. This Letter introduces a local ZigZag Max histograms of pooling pattern (LZMHPP) for classification of texture images. To compute the descriptor, first the dissimilarity between centre pixel and its neighbours is computed for each image patch over the whole image, then the dissimilarity map is encoded with different type of ZigZag ordering mechanism, and finally the Max histograms pooling is used to form LZMHPP descriptor from two complementary ZigZag weighted structures and achieves sufficient robustness under geometric variations. The experimental study on KTH‐TIPS and CUReT texture databases indicates the efficiency and supremacy of LZMHPP descriptor for texture classification.