Copy-Move Forgery Detection Using Cellular Automata - Cellular Automata in Image Processing and Geometry - page 109

as a binary image, but this must be done in a proper manner that ensures that enough
information is kept for further analysis.

6.4.1.1 Image Thresholding

A grayscale image can be represented in binary by simple thresholding of pixel
intensity values g i using a predefined threshold T b according to equation (6.4).

1

, g i ≥ t b

g i = (6.4)

0 , g i < t b

Threshold T b has to be selected according to image content with the aim of pre-
serving sufficient information about image textures so enough descriptive informa-
tion is kept for further analysis. Figure 6.4 shows a few examples of grayscale image
thresholding using different thresholds. The use of smaller values (for example, 50)
results in not enough details in the image, as can be seen in Fig. 6.4b. Larger val-
ues for thresholding (Fig. 6.4d) leads to loss of some details in the image. For this
example, a good threshold value would be around 70, as can be seen in Fig. 6.4c.
Even if using mean or median values for image thresholding appears to be an intu-
itive solution, global use of those values can lead to unreliable results (see Fig. 6.4e
and 6.4f) where most information of image content and texture is lost.
The use of binary images obtained by simple thresholding of grayscale images as
an input into CA is not the best solution for CMFD. A problem arises since applying
the same (global) threshold to all pixels values results in many homogeneous regions
in the binary image, especially in areas with smooth textures. A possible solution for
this problem is to generate more binary images with different thresholds, so-called
threshold decomposition [18], and use all of them as an input into the algorithm.
The final result is then accomplished by combining (summing) the results from all
binary images. Even if this approach can lead to good results, better results can be
accomplished by different binary representation.

6.4.1.2 Image Binary Planes

Another way of representing a greyscale image in a binary way is to convert all pixel
values into binary codes and use each plane of bits as one binary image. The result
of this process is 8 different binary images, as can be seen in Fig. 6.5). Naturally, the
use of the least significant bits will produce an image similar to noise (Fig. 6.5b),
but that effect drops with moving to more significant bits (Fig. 6.5f- 6.5i). Binary
images obtained by using more significant bit planes are more similar to images
accomplished by thresholding of the greyscale image, especially the image for the
most significant bit (Fig. 6.5i). According to that, the use of those images would
lead to the same problem as in the case when thresholding is used. To avoid that,
images obtained by using lower bits should be used. Even if they look like noise
(especially Fig. 6.5b- 6.5d), they contain a detailed description of texture. Also, the
lack of uniform white or black areas on those images leads to fewer falsely detected
areas.