Research on image quality evaluation and its application in image fusions

In view of the digital image in the process of acquisition, storage, processing and transmission due to the camera equipment, compression degree and transmission bandwidth caused by various distortion problems, this paper summarizes the image quality evaluation methods. Firstly, two methods of image quality evaluation, namely subjective evaluation and objective evaluation, are introduced, and then the relationship between subjective evaluation and objective evaluation is analyzed. Finally, the image quality evaluation method is successfully applied to the image fusion processing test, which provides ideas for the subsequent research on image quality evaluation methods.


Introduction
The quality of the image directly reflects the ability of the image to provide accurate information to people or equipment, which is related to the accuracy and sufficiency of the information obtained.Image quality not only depends on imaging equipment, imaging environment, illumination intensity, instrument noise, and so on, but also in a series of links of image processing, transmission and storage, will be affected by external factors, resulting in different degrees of image quality distortion, which brings great trouble to the later image processing.In recent years, with the rapid development of computer technology, the application of image processing technology is more and more extensive, image quality evaluation is becoming more and more important.Image quality evaluation has become one of the most important researches in the field of image information engineering [1,2].

Literature review
Image quality assessment (IQA) can be divided into Subjective image quality assessment (SIQA) and Objective image quality assessment (OIQA).The subjective assessment method is intuitive, accurate and effective.However, in a certain test environment, the image is scored by multiple observers.Then a large number of scoring data for statistical analysis, so time-consuming, not easy to achieve; Objective evaluation method is generally based on a mathematical model to calculate the score of image quality, relatively economical and practical, but it requires that the results of objective evaluation should have a good correlation with the results of subjective evaluation [3,4].

Subjective evaluation method
Subjective quality evaluation method refers to that the observer directly observes the image based on the visual reflection of human eyes, and then evaluates and scores the image quality according to the standards he believes.For a long time, many scholars believe that the subjective quality evaluation method is the most intuitive and commonly used in image quality evaluation.From a scientific point of view, the human eye is considered to be the most sophisticated imaging system known.Therefore, the subjective evaluation of the observer is characterized by high efficiency and simple operation, which can effectively reflect the visual quality of the restored image.However, due to the influence of external factors such as people's preferences, knowledge and culture level, and the environment during observation, the subjective evaluation method cannot obtain a unified standard.In addition, this method is not suitable for automatic monitoring of computers for another important reason.In practical application, the subjective quality evaluation method only takes part in the work of image quality evaluation in an auxiliary form.Table 1 shows the five quality scales commonly used in the world and their corresponding obstacle scales [5,6].
Table 1.Subjective evaluation results Very uncomfortable feeling Channel Theoretically, the most accurate and reliable evaluation method should be the subjective evaluation of image quality.However, from the above description, it is not difficult to draw the conclusion that the evaluation process will be very time-consuming due to the limited human energy and efficiency.It is necessary to build a specific evaluation environment, hire a certain number of evaluation personnel, and consume a lot of manpower and material resources.The evaluation result is easily affected by the working background of the evaluator, the environment of the evaluator and the emotion of the evaluator.All these will lead to repeatability and uncertainty of subjective evaluation methods.As a result, the image quality cannot be objectively reflected.Therefore, the subjective quality evaluation method cannot be used as the final evaluation, but only as a reference basis [7,8].

Objective evaluation method
In view of the uncertainty of subjective quality evaluation method, it requires people to explore an objective and quantitative evaluation system with theoretical support.Objective quality evaluation method is more important.Therefore, the commonly mentioned image quality evaluation method refers to the objective evaluation method, whose goal is to obtain the objective evaluation value that can be consistent with the subjective evaluation results [9,10].
Objective evaluation is to select the appropriate mathematical statistical algorithm according to the characteristics of the image, and finally evaluate the quality of the image quantitatively in the form of data.To a certain extent, using data to evaluate image quality has a better explanation than subjective image quality evaluation.With the support of data, the credibility will be improved, and the impact of environment will be reduced.However, this method also has drawbacks.Firstly, the data obtained through calculation are usually scattered; secondly, it is very vulnerable to external influences during the imaging process, which will eventually lead to the accuracy of the image value.In addition, the definition of objective quality assessment is too broad, and does not restrict the algorithm or the original image type, which lacks certain pertinence.Usually, the objective quality evaluation methods are as follows: mean squard error (MSE), peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), average structural similarity index measure (SSSIM), in addition, there are signal-tonoise ratio (SNR), mean gradient and other evaluation methods.PSNR and SSIM are the most common and widely used in practice [11,12].

Mean Squared Error
The mean square error is applicable to the simulation of the known original image and is used to compare the deviation between the simulation result and the original image.The smaller the mean square error, the better [13,14].
Set  � , 、 .represent the restored image and the original image in(i, j)location of pixel values, M、N is the number of columns image.The specific calculation formula is as follows: (1)

Peak signal to noise ratio
Peak signal-to-noise ratio (PSNR) is the most widely used objective method to evaluate image quality.In this method, the recovered image is regarded as the original image mixed with noise.The larger the tested value is, the less the image distortion is, the closer the restored image is to the original image, that is to say, the quality of the restored image is the best.The larger the peak signal-to-noise ratio, the better.The specific formula is as follows: (2) where L represents the maximum gray value of the image.For 8-bit images, L is 255.Both MSE and PSNR are based on statistical features, which are easy to calculate and very intuitive [15].

Structural similarity index measure
Structural similarity measurement is a method to measure the similarity between the image to be evaluated and the original image based on structural information.The larger the value, the better, and the maximum value is 1 (the original image and the reconstructed image are 100% identical).This method is easy to calculate, especially consistent with the subjective perception of human eyes.The specific calculation formula is as follows: Where () and � ̂� respectively image  and  ̂ averages,   2 and   ̂2 respectively image  and image  ̂ variance,   ̂ said image  and image  ̂ estimate covariance [16].On the surface, there seems to be no connection between the subjective evaluation method and the objective evaluation method.However, after practical research, there is a certain connection between the two evaluation methods, which can be roughly converted from each other.The relationship is shown in Table 2 [17,18].
Table 2.The relationship between the two evaluation methods

Objective evaluation criteria (peak signal to noise ratio)
You don't feel the distortion PSNR>48 You felt distorted, but there was no discomfort 35<PSNR≤48 You feel slightly ill 29<PSNR≤35 An uncomfortable feeling 25<PSNR≤29 Very uncomfortable feeling PSNR≤25 Therefore, a conclusion can be drawn that whether the subjective quality assessment method or the objective quality assessment method is selected, the specific environment and the type of the selected algorithm should be combined with careful analysis, so as to choose a suitable evaluation method.
Among these three common objective quality assessment methods, PSNR and SSIM methods are chosen in this paper for the following reasons: MSE is often different from people's subjective feelings, and has low correlation; PSNR is not only simple and intuitive, but also related to the gray level of the image.The most important thing is that it is consistent with people's subjective feeling.Usually, the image quality is directly proportional to PSNR, so the peak signal-to-noise ratio becomes the standard for evaluating image quality in practical applications.SSIM is associated with image features, so it is also an indispensable evaluation method.

Formatting the text
In this paper, multi-spectral images (Fig. 1) and SAR images (Fig. 2) are selected as test images to verify the effectiveness of fusion algorithms such as weighted average fusion, Brovey transform fusion, PCA transform fusion, HIS transform fusion, à trous wavelet transform fusion and so on.The fusion results are shown in Fig. 3 to Fig. 8 respectively.

Subjective evaluation
As can be seen from the above figures, the fusion results obtained by these six fusion methods can retain the spectral information in the optical image well and improve the image resolution.Table 3 shows the visual effect evaluation results of the fusion images obtained by these six image fusion methods [19].

Objective evaluation
Evaluation standards such as PSNR and SSIM were used to evaluate the processing results of the obtained fusion result graph, as shown in Table 4, so as to compare the correlation degree between the fusion effect of the judgment algorithm and the original high-resolution image [20][21][22].The peak signal-to-noise ratio reflects the degree of distortion between the fusion image and the original image.As can be seen from the data in the table above, the peak signal-to-noise ratio of wavelet transform fusion method is 55, which is the maximum value.Therefore, the degree of distortion after fusion of this algorithm is small.The structural similarity measure reflects the similarity degree between the fusion image and the original image.It can be seen from the data in the above table that the structural similarity measure of the wavelet transform fusion method is 0.9988, which is the largest value.Therefore, the image fusion algorithm has a high similarity with the original image.Comprehensive image subjective quality evaluation and objective quality evaluation, due to the latter two algorithms of PSNR>48.We don't feel the image distortion.However, the fusion algorithm based on a trous wavelet and HIS transform has high similarity, so the fusion effect of this algorithm is better.

Conclusion
Image quality evaluation is a research hotspot in the field of image processing.The ideal image quality evaluation index should meet the following three aspects: first, the image evaluation results have a good agreement with the direct feeling of human vision; Second, the image evaluation index is universal, which can be applied to a variety of image processing technology and a variety of fields; Thirdly, the results of image quality evaluation are monotonous, accurate and consistent.In this paper, two methods of subjective quality evaluation and objective quality evaluation are sorted out.After fusion processing of multi-spectral image and SAR image by several fusion algorithms, the image quality evaluation is carried out by combining subjective quality evaluation and objective quality evaluation.The experimental results show that the subjective quality evaluation is basically consistent with the objective quality evaluation, and the accuracy of the fusion algorithm is verified, which provides support for the subsequent research on image quality evaluation methods.

Table 3 .
Subjective evaluation results of six fusion methods

Table 4 .
Objective evaluation results of six fusion methods