现在的位置: 首页 > 综合 > 正文

GLCM问题记录

2013年11月26日 ⁄ 综合 ⁄ 共 2253字 ⁄ 字号 评论关闭

Texture Features: Texture contains importantinformation in image classification, as it represents the contentof many real-world images. Textures are characteristic intensity(or color) variations that typically originate from roughness ofobject
surfaces (Davies, 2008). As a powerful source ofinformation, texture features have been intensively studied inremote sensing image classification (Zhang et al., 2004,Franklin et al., 2000, Reulke et al., 2005, Samal et al., 2006).There are many different methods
used to extract model texturefrom images. In this paper, we evaluated the widely used GLCM texture measures and state-of-art texture descriptor Local Binary Patterns (LBP) and its extensions: uniform LBP, rotation-invariant LBP, dominant local binary patterns
(DLBP).In this section, an overview of these texture descriptors is given.

The image-objects generated from segmentation is arbitraryshaped,however, texture measurements are usually extracted based on the texture property of pixels or small blocks withinthe rectangular shaped region. Therefore, in this paper, the arbitrary-shaped
objects are extended to a rectangular area fortexture extraction. This can be achieved by padding zero ormean value outside the object boundary, or obtaining the innerrectangle from the object. Zero padding introduces spurioushigh frequency components leading
to degrading the performance of the texture feature, while the inner rectanglecannot usually represent the property of the entire object well. Mean-intensity padding has shown better performance than theother two approaches (Liu et al., 2006) and thus is adopted
inthis paper. Firstly, the minimum bounding rectangle is obtainedfrom the image segment, and then the area which is outside of the segment and inside of the minimum bounding rectangle ispadded using the mean value of pixels in the region.

Grey-level co-occurrence matrices (GLCM) have beensuccessfully used for deriving texture measures from images.This technique uses a spatial co-occurrence matrix thatcomputes the relationships of pixel values and uses these valuesto compute the second-order
statistics (Haralick et al., 1973).The GLCM approach assumes that the texture information in animage is constrained in the overall or “average” spatialrelationships between pixels of different grey level. In thispaper, we use mean and standard deviation of
four measuresfrom the grey-level co-occurrence matrices: energy, entropy,contrast, and homogeneity.

摘自《EVALUATION OF SPECTRAL AND TEXTURE FEATURES FOR OBJECT-BASED VEGETATION SPECIES CLASSIFICATION USING SUPPORT VECTOR MACHINES》

抱歉!评论已关闭.