140.

[Cover] [Contents] [Index]

Page 223

of the use of geostatistical methods for the characterisation of image texture are provided by Carr (1999) and van der Meer (1999b).

5.6 Experimental analysis

5.6.1 Test image generation

Brodatz’s texture images (Brodatz, 1966) are widely used for texture analysis comparisons. Here, we employ a combination of six Brodatz texture patterns (Figure 5.21) to test the performance of the different texture extraction approaches described in the preceding sections. Each subimage in Figure 5.21 is 128×128 pixels in size. From visual observation, the differences between the texture patterns within the image are clearly apparent. Experiments were also performed on a set of images subjected to an independent identical noise-fading environment. The original images were corrupted with additive noise denoted as:

(5.54)

where x(i, j) is the original pixel value at location (i, j), e(i, j) is the independent identical distribution (i.i.d.) Gaussian noise, and g(i, j) is the resulting noise-fading image. The effect of noise-fading is determined by the signal to noise ratio (SNR), defined as the ratio of average energy of the original image to the average energy of the random noise (Won and Derin, 1992):

(5.55)

Let v(i, j) be a normal-distributed random variable at (i, j), then e(i, j)= Kv(i, j), where:

(5.56)

Once the SNR is determined, Equation (5.56) is applied to obtain the values of K and then e. Equation (5.54) is then applied to generate the corresponding noise-fading image.

Four images were generated based on Figure 5.21a with different SNR

[Cover] [Contents] [Index]