No-reference Image Quality Assessment Based On Empirical Mode Decomposition

In this Internet era of information sharing,people are exposed to different forms of information every day.As the source of visual information,images contain the advantages of intuition and accuracy that cannot be matched by other types of information.The final presentation of the image needs to go through many processing steps in people's sight,resulting in image distortion and quality degradation.Therefore,the assessment of image quality is a very important part in obtaining image information efficiently,and has aroused the scientific research interest of researchers in the field of image processing.At this stage,the objective image quality assessment methods can be divided into three categories according to the degree of relying on the reference image information:full reference(FR),reduced-reference(RR),and no-reference(NR)image quality assessment(IQA).The reason why the study of no-referenced image assessment method becomes a hot topic is that the original image cannot be obtained under normal circumstances,so the study of no-reference image quality assessment method is more in line with the actual situation.This paper proposes two no-reference image quality assessment algorithms based on empirical mode decomposition and applies them to defect identification of solar photovoltaic cell.The specific research contents are as follows:(1)The SSEQ algorithm proposed by Liu et al is improved,Which is based on the local spatial entropy and spectrum entropy features of images.And its spatial information entropy is changed into gradient information entropy.The EEMD algorithm combining information entropy and empirical mode decomposition is proposed.Firstly,the first two intrinsic mode functions were obtained by empirical mode decomposition of the image,and the gray values of the original image and the first two intrinsic mode functions were reordered in order.The mean and skewness of the gradient information entropy and the mean and skewness of the spectrum information entropy are extracted from the sorted original image and the sorted first two intrinsic mode functions.Then,the sorted original image and the sorted first two intrinsic mode functions are convolved with Scharr operator and LOG operator in turn to extract the mean value and skewness of the spectrum information entropy and the statistical characteristics of distribution.Finally,the model of no-reference image quality assessment was built by support vector machine and the training test experiment was performed on the LIVE image database.The results show that the EEMD algorithm improves the SSEQ algorithm to obtain higher quality assessment accuracy.The median value of SROCC value is 0.9442,the median value of PLCC value is 0.9500,the median value of RMSE value is 7.2068,and the standard deviation of SROCC value,PLCC value and RMSE value is 0.0197,0.0189 and 1.0056,respectively.The image classification accuracy is 87.89%,which is better than the four mainstream image quality assessment algorithms of PSNR,BIQI,BLIINDS-? and SSEQ.(2)The EMDMFF algorithm by multi-feature fusion based on empirical mode decomposition is proposed.The algorithm firstly performs empirical mode decomposition on the image to obtain the first two intrinsic mode function,and uses the fractional differential operator to enthance it,three features are extracted from the enhanced original image and the first two intrinsic mode functions after enhancement.Three features above:the first is to extract the statistical characteristics fitted by the brightness normalization coefficient,the second is to extract the gradient-weighted LBP histogram features,three is to extract the structural similarity features;Then,the no-reference image quality assessment model is built by using support vector machine.Finally,1000 iteration training test experiment was performed on the LIVE image database.The results show that the EMDMFF algorithm adopts the method of multi-mode and multi-feature fusion,which makes the image quality assessment algorithm have higher accuracy.The median value of SROCC value is 0.9568,the median value of PLCC value is 0.9575,the median value of RMSE value is 6.6221,the standard deviation of SROCC value,PLCC value and RMSE value is 0.0062,0.0074 and 0.5877,respectively.The image classification accuracy rate is 92.62%,which is better than 8 mainstream image quality assessment algorithms such as BRISQUE,BIQA,SSEQ and WPD.Compared with these 8 algorithms,EMDMFF algorithm is competitive to some extent and more consistent in subjective quality perception.(3)The EEMD algorithm and EMDMFF algorithm are used to identify whether the solar photovoltaic cells images is defective and the type of the defect.20 non-defective and 382 defective solar photovoltaic cell images are constructed into a solar photovoltaic cell image database similar to the LIVE image database,and the photovoltaic cell image database is trained and tested using EEMD algorithm and EMDMFF algorithm to obtain image defect classification.The recognition results are 70.24%and 72.19%respectively.After performance comparison,it is found EMDMFF algorithm has the certain application value in the defect recognition of solar photovoltaic cells,and the image quality score obtained from the assessment can be used to determine the defect level of the image.