An attempt was made to trace the possibility of division of the content of satellite images into two basic classes associated with the "low" and "high" values of the texture. This classification approach is applied during object-oriented classification and results are dependent on spatial resolution. On the basis of panchromatic channel of KOMPSAT-2 image of 1m resolution a data set with a resolution of 1, 2, 4, 8, 16, 32 and 64 m were prepared. Then images were processed using selected texture functions: Sobel, Laplacian and Sigma filters, transformation PanBF as well as Haralick functions: correlation, homogeneity and entropy. On the basis of texture images an analysis of discrimination of four basic land cover classes has been done: built-up areas, forests, agriculture areas and water. These classes were selected because built-up areas and forest belong to “high” texture and remaining two are usually represented by “low” values of texture. For each texture image form using different functions and spatial resolution, Bhattacharya distance and next Jeffries-Matusita (J-M) distance between land cover classes were calculated. Results are presented in tables 1, 2, 3, 4, 5, 6 and 7. They also include J-M distance between “low” and “high” texture. It was assumed that two classes are well seperated if the value of J-M distance is over 1.7. In the case of all seven texture transformations the best results of class discrimination were observed for images with the highest resolution. Distinct deterioration of discrimination between “low” and “high” texture took place in the case of images with a resolution of 8 m or less. By far the highest J-M values were obtained on the basis of Laplacian filter and next using Sigma filter, PanBF and the Sobel filter. In comparison usefulness of the Haralick function has proved much less. The presented results could be practically applied in the work on classification algorithms of very high, high and medium resolution satellite images.