Window size for texture analysis

Hi all,

I am doing feature extraction for texture analysis now. I am using a moving window kernel to derive new variable images such as first-order derivative images and GLCM with several different windows size (7x7 pixels, 11x11 pixels, 19x19 pixels, 31x31 pixels). Although, visually, the differences can be noticed, is there any reference how to choose the best kernel size for this. The purpose of this work is to classify/detect the objects with min 0.5 meter of width size and the pixel image resolution is 5cm. Thank you in advance.

Answer 1

[N/A]

Quote from "Mapping invasive Fallopia japonica by combined spectral, spatial, and temporal analysis of digital orthophotos" by Wouter Dorigo et al.:

"In our study, we established the optimum kernel size in an iterative way, by evaluating the classification performance of GLCM measures based on increasingly large kernel sizes, as proposed by Murray et al. (2010)."