Objective Image Quality Assessment Using MATLAB-based Algorithms

Objective Image Quality Assessment

1. Introduction 

Objective IQA is a technique that uses mathematical models to predict the image quality. Contrary to Subjective IQA, Objective IQA is automatic and accurate hence giving better results. Consequently, it is widely applied in various fields. For instance, in control systems to monitor image quality. Such systems adjust themselves automatically resulting to best quality image data. Additionally, objective IQA can be used to select among alternatives the best algorithm that gives higher quality images (Mohammadi, Moghadam and Shiran, 2014).

Mean Square Error (MSE)

MSE is a measure of the signal fidelity by comparing two signals. Fidelity measure provides the quantitative score which measures the level of similarity or the degree of distortion between the two images. If x and y represent the original and degraded signals respectively then, the measure of signal quality, MSE can be given as . 

                                                                                         

Often, the MSE can be converted to PSNR by the relation, . 

                                                                                           

Where,  -1 refers to the range of the acceptable pixel intensities whose value is 255. PSNR is measured in decibels and from the relation, we can deduce that it is inversely proportional to MSE.

Average Difference(AD)

“Average difference refers to the average of the difference between the original image and the degraded image. It is given by the equation” (WEI et al., 2013).:

                                                                                      

Maximum Difference (MD)

“Refers to the maximum difference between the reference image and the distorted image” (WEI et al., 2013).

                                                                                   

Objective IQA Metrics

Normalized Cross-Correlation (NK)

NK refers to the closeness to which the reference image and the degraded images are quantified.

                                                                                    

Structural content (SC)

“It is a measure of the similarity between a reference image and the distorted image” (WEI et al.,2013). Where,  and  represents the original image and the distorted image respectively”.

                                                                                 

a) Problem Statement

Image processing is applied in various fields. Therefore, there is the need to measure the quality of the images used. The photos could be degraded due to physical limitations from the time they were captured to the time they are humans view them. Therefore, knowing such distortions could help designers to code and develop systems that have the highest sensitivity to these distortions.  Notably, subjective evaluation is used to quantify visual image quality. However, such methods are expensive, time-consuming and lack automation. On the contrary, objective computational metrics are automatic and can measure image quality and record the results without human intervention. Objective IQA could eliminate the need for inconvenient, expensive, and time consuming subjective image quality assessment means.

b)Objective

The main aim of the lab assignment is to design and simulate “objective Image Quality Assessment” methods using MATLAB-based algorithms. The lab also focuses on using two test images to carry out the lab and note the results to be obtained.

c)Limitations

Despite the fact that MSE and PSNR are used to measure the image quality, the methods are still susceptible to energy of errors.

Requirements

MATLAB 2016 and two images (the original image and the degraded image) will be used in the lab assignment.

 2. METHODOLOGY

MATLAB 2016 software has designed the various image quality assessment metrics. MATLAB software possesses “matrix handling capabilities” and excellent graphics. Additionally, MATLAB provides a powerful inbuilt toolbox thus offering a conducive environment for technical computing. Most importantly, it has a “separate toolbox for image processing applications” (Mohammadi, Moghadam and Shiran, 2014).

Design Procedure

Step 1: Study the metrics already developed for measuring image quality.

MATLAB-based Algorithm Design Methodology

Step 2: Select the original image and corrupt it with some noise to obtain the distorted form of the image.

Step 3: Develop the algorithms and simulate the methods

Step 4: Analyze the results obtained and deduce conclusions based on the analysis.

3. MATLAB DESIGN, RESULTS AND ANALYSIS

a) The images considered were image1_Original.jpg and image2_Distorted.jpg where the latter represents the degraded image. Both images are of the size 512 by 512 pixels.     

                                            

The results obtained are shown in table 1

Table 1: MATLAB results

Assessment Method	Value Obtained
“Mean Square Error (MSE)”	7.17
“Peak Signal to Noise Ratio (PSNR)”	39.6098815
“Normalized Absolute Error (NAE)”	0.01
“Maximum Difference (MD)”	220
Structural Content (SC)	1.01
Average Difference (AD)	1.7
Normalized Cross Correlation (NCC)	1.00

                         

Figure 3: MSE Output

                        

Figure 4: PSNR Output

                         

Figure 5: NAE Output

                                   

Figure 6: MD Output

                                    

Figure 7: SC Output

                                             

Figure 8: AD Output

                                         

Figure 9: NCC Output

c) “Importance of objective assessment in image processing”

Over the recent past, the demand for digital image-based applications has seen considerable growth in all sectors of the economy. For instance, there have been widespread image processing applications ranging from the medical research to industrial applications. Often, these applications require high-quality image processing techniques as required by human quality judgments. As a result, the efficiency and reliability of image quality evaluation mechanisms have become fundamental (WEI et al., 2013). Therefore, Image Quality Assessment (IQA) can be done by either subjective quality assessment or through objective quality assessment.

Objective Image Quality Assessment methods use mathematical models to predict as well as measure the image quality accurately and automatically. As such, an ideal model mimics the expected quality levels of an average human. The conventional objective method employed is full reference IQA where the original, perfect image is used as a reference. Additionally, in reduced reference IQA the undistorted original image is partially available. Also, objective IQA employs no-reference IQA when the reference image is unavailable. Objective quality assessment methods are widely preferred than subjective processes due to many strengths.

Results and Analysis

First, objective IQA, as opposed to subjective IQA methods, are less expensive and simple to calculate with less computational complexities. Since they are software-based, a given algorithm can be implemented to simplify all the computational works. Hence it becomes smooth and faster to carry out evaluations to ascertain the image quality. Mainly, MSE and PSNR are sensitive to varied types of distortions.

Secondly, these methods can offer the “repeatability and reliability” missing in subjective image evaluations. “A machine-vision-based system can provide detailed information about individual attributes that count to the overall perception of image quality” (WEI et al., 2013). Such a system not only characterizes line-quality and dot quality, but it also describes color reproduction and other details. It is imperative to maintain a close correlation between the machine attributes and observer-based response. However, an objective assessment system provides additional information that can be used to evaluate various causes of a given defect (Mohammadi, Moghadam and Shiran, 2014).

Additionally, the methods provide for automation thereby boosting the capabilities and benefits significantly. An automated system that is designed well can handle assessment of a large volume of images as well as many image attributes. Therefore, such a system is critical in failure analysis and control of statistical processes.

4. CONCLUSION

Image quality measurement is paramount in various applications. “In the recent past, efforts have been made to develop objective image quality evaluation techniques and algorithms”(Shanableh, 2015). As a result, the lab instilled an understanding of the various objective methods of Image Quality Assessment and how they can be modeled in MATLAB.

5. Bibliography

 WEI, J., LI, S., LIU, W. and ZANG, Y. (2013). Objective quality evaluation method of stereo image based on steerable pyramid. Journal of Computer Applications, 32(3), pp.710-714.

Subjective and Objective Quality Assessment of Image: A Survey

Mohammadi, P., Moghadam, A. and Shiran, S. (2014). Subjective and Objective Quality Assessment of Image: A Survey. Subjective and Objective Quality Assessment of Image, 57(3), pp.1-51.

Shanableh, T. (2015). A regression-based framework for estimating the objective quality of HEVC coding units and video frames. Signal Processing: Image Communication, 34, pp.22-31.

Hooker, D. (1973). Retrieved from https://en.wikipedia.org/wiki/Lenna#/media/File:Lenna.png

6. 

Average Difference (AD)

                                    

Maximum Difference (MD)

                                  

Mean Square Error (MSE)

                                

Normalized Absolute Error (NAE)

                               

Normalized Cross Correlation (NCC)

                               

Peak Signal to Noise Ratio (PSNR)

                               

Structural content (SC)