Image Quality Assessment Based on SIFT and SSIM - Advances in Image and Graphics Technologies - page 5

3 Experiments Analysis and Comparison

To verify the performance of the algorithm, were carried out tests various distortion
and overall tests on a public database of LIVE database2 [9]. Basic information of
LIVE database is shown in Table 1.

Table 1. Basic Information of LIVE Image Database

Information/
Database

Number of Refer-
ence Images

Numbers of Dis-
torted Images

Number of
Distortion

Image
Type

LIVE 29 779 5 RGB

Gaussian Blur
JPEG
JP2K
White Noise
Fast Fading

Distortion Type

Contrast algorithms include five classic full reference algorithms: PSNR [2], SSIM [3],
MS-SSIM [4], FSIM [5], and RFSIM [6]. Index of comparison Algorithm is Spearman
rank correlation coefficient (SROCC) and Pearson correlation coefficient (PLCC).
To find suitable threshold of quality assessment correlation, we range the Ratio
from 0.1 to 0.9, and even form 0.91 to 0.99, then SROCC is calculated, which is
shown in Table 2. The results show that, for the assessment of the role of the distor-
tion effects in the two images, the high value of Ratio maybe helpful to enhance
matching feature points. Therefore, the paper takes Ratio to 0.99.

Table 2. Relationship of Ratio and SROCC in FF

Ra-
tio

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.91

0.795
4

0.859
3

0.904
2

0.927
4

0.940
6

0.947
6

0.951
7

0.950
4

0.947
6

Ra-
tio

0.92 0.93 0.94 0.95 0.96 0.97 0.98 0.99

0.945
8

0.945
7

0.945
2

0.944
7

0.945
1

0.948
5

0.950
9

0.952
2

The assessment results of various algorithms are shown in Tables 3 and 4.

Table 3. SROCC

Algorithm\Distortion JP2K JPEG WN Blur FF ALL

PSNR 0.9119 0.8774 0.9365 0.7669 0.8869 0.8759

SSIM 0.9370 0.8974 0.9099 0.9065 0.9396 0.9181

MS-SSIM 0.9370 0.8983 0.9213 0.9344 0.9350 0.9252

FSIM 0.9386 0.8965 0.9189 0.9527 0.9515 0.9316

RFSIM 0.9275 0.8890 0.8914 0.9232 0.9120

0.9291

SIFT-SSIM 0.9378 0.9060 0.9179 0.9298 0.9522 0.9287