The 2D Gabor filtering mainly captured information based on Scales and Orientations. Therefore the main criteria would be the texture's scales (or size) and orientations.

Generally, you need to test all possible reasonable scales and frequencies. Some people say it is depend on the textures within your images. But it is really important to test all the possible orientations and scales then test them individually and find out which ones give you the best results. In some cases combinations of different orientations/scales (e.g. by taking the average) give you better results. However, if you just want to make a primary investigation, for scale selection you may want to consider the size of your textures. For example small textures can be captured with scales which are slightly bigger than your texture size. You can't (you can but it is less appropriate) capture it using a scale which is too large from the size of your texture because you may over-smooth or took unnecessary information from your textures.  In many cases textures can be represented in different scales in an image. In this case you need to use different scales to capture these textures.

@Andrik Rampun  Thank You

Hi Prem,

Here's a paper you might find useful:

Bianconi, F., Fernández, A. Evaluation of the effects of Gabor filter parameters on texture classification (2007) Pattern Recognition, 40 (12), pp. 3325-3335.

Apologies for citing my own work ;-)

Best,

f.

Hi, Dr. @Andrik and Dr. @Francesco. Recently, i am doing a research work which is related to 2-D Gabor filters. I am confused about the range of lambda(i.e. the derivative of frequency). What's lower bound of lambda ? Some study claims that lambda is usually greater than 2. However, some blogs said that only if lambda greater than 0 is enough, the specific value depends on different applications.