Page 97 - The-5th-MCAIT2021-eProceeding

((2 )−1) 2
= 10 (3)

Where MSE represents the Mean Squared Error, which is the average value of the combined square of errors,
or differences in each pixel between the original and compressed image (Sharma 2017). The equation to
measure MSE is as illustrated as follows:
1
2
= ∑ ( − ̂ ) (4)

=1
However, there are certain instances in which the MSE value of an image becomes inconsistent and hence a
secondary measure of image quality, SSIM is introduced. SSIM measures image quality degradation between
the original and compressed image by observing the perceivable structural differences between the two images
and serves as an alternative besides PSNR for image quality measurement (Wang 2004). The equation to
measure SSIM is as illustrated as follows with μ x, μy representing the average of x and y, and σ x, σy representing
the variance of x and y, and σ xy representing the covariance of x and y:
(2 + 1 )(2 + 2 )
( , ) = 2 2 2 2 (5)
( + + 1 )( + + 2 )
3. Findings

After compression has been applied on the dataset, the Encoding Time, CR BPP , PSNR and
SSIM) for each image using each algorithm was recorded in a table. The average value for these parameters
were then calculated and tabulated in the table presented below.

Table 1. Average compression results for each algorithm on a sample of 85 images

From the table above, we can observe that the DWT algorithm produces the lowest performance out of all
four algorithms that were tested. While the other three each excels in certain criteria. DCT produces a fast and
a good quality compression, as seen with high PSNR and SSIM numbers, but with a lower CR compared to
Huffman and SPIHT. Huffman on the other hand, is also fast and produces a much more compact compression
as seen in the high CR numbers but produces a lower quality compression when compared to DCT and SPIHT.
Amongst all the algorithms that were tested, SPIHT produced a very good overall result, being quite similar in
performance with the top performing algorithm in each aspect, resulting in a high CR, PSNR and SSIM
compression performance, with only one caveat, and that is the amount of time taken due to the complexity of
the algorithm resulting in slow encoding speeds.

4. Conclusion

From the data that was gathered through this research, it can be clearly seen that image compression serves
as an effective way to reduce the size of images, whether it be for more efficient storage, or for increased transfer
efficiency. Based on the results, it can be said that the SPIHT algorithm performs well in all aspects besides
encoding time. This can be a point to focus on in coming research to optimize the efficiency of the SPIHT

E- Proceedings of The 5th International Multi-Conference on Artificial Intelligence Technology (MCAIT 2021) [84]
Artificial Intelligence in the 4th Industrial Revolution