Page 86 - IJEEE-2023-Vol19-ISSUE-1
P. 86

82 |                                                                               Hussein, Muhajjar, & Mahdi

                 IV. IMPLEMENTED MEASURES

  A. Mean square error (MSE)

     It shows the exact square difference between the un-
watermarked and watermarked versions of the image. In
general, MSE has no precise value, although the lower the
value, the better, and zero are optimal. The MSE define as
equation1[17]:

      MSE =  1  M´N    (Ci - Si )2           (1)
                  å
             M ´N i=0

while M×N is the size of the image, Ci and Si are,
respectively, watermark and host images. a main problem
with MSE it depends to intensity of pixel.

  B. Imperceptibility                                          (a) Original image  (b) watermarked image

Regarding watermarks, imperceptibility is one of the most      Fig. 2: TEST1 original image and the watermarked image
important aspects to consider. As a result, the amount of
noise introduced to the picture must be quantified by the                                       TABLE I
watermark bits to evaluate the quality of the reconstructed    THE RESULT AFTER AN ATTACK ON AN IMAGE WITH A FRAGILE
image. This unit of measurement is referred to as an image
quality metric. The Peak Signal to Noise Ratio (PSNR)                                        WATERMARK.
define as equation (2) [18]

PSNR = 20 ´ log10 (MAXl) - 10 ´ log10 (MSE)  (2)

Max is the highest pixel value that may be achieved, while
MSE is the mean square error. .it important to know that
increasing the ratio of PSNR will improve image quality,
which is the most important factor in the proposed system

                  V. RESULTS OF EXPERIMENT                      TEST1 image         TEST2 image
                                                               PSNR=70.83db        PSNR=69.03db
The supplied form has been executed on a dataset taken from
the Kaggle website as part of this proposal [19]. Table I       MSE =0.003          MSE =0.008
represents this dataset and the result after embedding the
watermark with the measures. Figure 2 shows the original        TEST3 image         TEST4 image
image and the watermarked image, and Figs. 3,4,5,6 will        PSNR=69.08db        PSNR=67.26db
include models with a histogram for the original and
embedded image drawn by python 3.9. As shown, there is no       MSE =0.008           MSE =0.01
difference between the histogram of the original image and
after the embedding process, which proves that the image
quality was not affected by inserting the watermark. And that
is the power of our work. The fragile watermark was
employed using an earlier technique for generating the secret
bits to authenticate the image and identify any manipulation.
   81   82   83   84   85   86   87   88   89   90   91