Blind Image Quality Assessment For Highly Distorted Images Using Ciqe Algorithm

DOI : 10.17577/IJERTV2IS3474

Download Full-Text PDF Cite this Publication

Text Only Version

Blind Image Quality Assessment For Highly Distorted Images Using Ciqe Algorithm

Anna Geomi. George1,a Kethsy Prabavathy2

PG Scholar/Dept.of.CSE1, Assistant Professor/Dept.of.CSE2


No Reference-methods are used to predict the Quality of the images automatically by extracting and modelling prior Knowledge on specific distortions where the subjective quality assessment requires But many of the times the color accuracy is not noticed ie the structural distortions. The statistical features have been changed due to the presence of distortions and can be identified eventually without any reference. In this paper structural and non structural distortions can be distinguished and identified using the CIQE (Color Image Quality Estimating Algorithm).Compare it with PSNR and SSIM method.

Keywords: NR Method, Objective QA,.Spatial Dispersion, Natural scene statistics(NSS) method, Human Visual System (HVS)


    Image quality means the measurement of the perceived image degradation and the factors affecting the quality of an image is sharpness, distortions,color accuracy,noise,artifacts etc.The quality assessment is one problem and is classified mainly in two, subjective and objective quality assessment. In subjective method the quality is measured with the help of human opinion score but it is time consuming one. so the objective method is introduced and it predict the quality automaticaly.Depending upon the original image as reference the objective method is classified in two three, Full reference, Reduced reference and No reference method[1]. These methods want to be highly correlated with human perception. Distortions are occurred during the transmission, storing and sharing of information.

    Many of the times only the distorted image is available and not have the original image as reference. In such cases the No reference method is used and it is difficult to create an algorithm to predict the visual quality without any reference. The Objective IQA measures are used in network visual communication applications for the purpose of QOS Monitoring. The concept of Human visual system is taken as it doesnt requires any reference to define the quality of an image. The statistical features also changes with distortions so by extracting the computational features from the image the distortions can be identified. The review

    and control of the quality of digital photographs is becoming quite challenging. By using the updating Algorithm the image quality can be evaluated and measured. In medical imaging application image distortions affect the diagnostic values so the quality want to be evaluate. These distortions can be identified by comparing the statistical features which are extracted from the input image with existing quality score. Then the type of distortion can be identified from it. The accurate and more efficient IQA measures will certainly enhance their applicability in real-world applications


    The No reference methods specify the distortions but it still lags the full reference method. The human eyes extract structural information from the viewing field, so the human visual system is highly adapted for this purpose. Therefore, the measurement of structural distortion should be a good approximation of perceived image distortion

    [2] from my survey.NR-IQA can be done based on DCT statistics. By this the BLIINDS algorithm is developed and predict the image quality based on the statistics of local discrete cosine transform coefficients[3]The degree of peakness and tail weight is then quantified.

    Where is the mean of x and is the standard deviation from this global image kurtosis value

    .Visually significant blocking artefact metric (VSBAM) is the other one used to measure the visibility of distortion as combination of blocking artefact and undistorted image.[4]But the LSB patterns are not satisfied in it. In BRISQUE method[5] the statistical features are extracted and the distortion orientation information is obtained by normalized luminance signals where in BLIINDS II the features are extracted in DCT domain and requires low dimensional features but it is time

    consuming one because of the computational block statistics of DCT[6]

    Image processing system

    Image processing system

    Updating algorithms

    Updating algorithms


    The input image is decomposed in using scale space orientation decomposition and extract the





    Quality measures

    Fig.1. Diagram for IQA

    In our method the quality is measured using the No Reference approach and the color accuracy is also considered.

    In NR- IQA method signal from the transmitter is digitized and processed at the front end and at the receiver the perceptual signal is taken and map the vectors and d display .The diagram is shown in fig.2.

    The psychophysical measurements to compute the visual quality and the image is decomposed to obtain the gain control model in the sub band decomposed domain[1] MSE (Mean Square Error) is used to evaluate the quality and is defined as

    MSE= –

    Where x is the original image and y is the distorted image N are the width and height of the image. When MSE value increases as the compression ratio increases. If the MSE value decreases to zero then pixel by pixel matching of images become perfect.MSE is a simpler one. The other one is PSNR and is denoted by

    PSNR=10 lo

    The MSE and PSNR methods are easy to implement and have low computational complexities but they are not highly correlated with human perception and fail to capture image quality when they are used to measure across distortion types.


    A. Natural scene statistics based feature extraction.

    statistical features using this the type of distortions can be identified. Range of pixel intensity values can be changed by normalization and the normalization can be done by the formula

    =(ne ) +

    Where is the width and is the intensity range. Divisive normalization affects the statistics of sub band coefficients.

    Figure 2.1 . 1) Jpeg and 2) white noise affect

    Figure 2.2. 1) Fast fading and 2) blur affect

    The natural scene statistics model is used for the evaluation of quality. These natural images are considered as the signals with certain statistical properties. These features capture the dependencies

    between the sub band coefficients over the scale and orientation statistics[7].

    B.Quality assessment

    Windowed structural correlation is used for comparing the band pass sub band and high pass residual band and is defined as[1]


    Where is the cross variance between BP and

    Hp bands, is the constant and are the windowed variances

      1. Histogram before normalization

        The blurring affect is due to the attenuation of the high spatial frequency coefficient which occurs during the compression stage and is denoted as:


        SVM based distortion identifier

        SVM based distortion identifier

        Different distortions quality predictor

        Different distortions quality predictor

        Where (i,j) is the pixel (i,j) intensity of a component of size the binary image resulting from the edge is the complementary.

        Input signal

        Fig .4 Distortion Predictor

        Spatial information and pixel activity is also included in it [8]. Blocking artefacts have been observed in block-based DCT compressed images

        .Obtain the feature vecor of it and by using the regression modules the quality is measured.SVM and SVR are mainly used for this.[9].SVM is used for the classification and contain the distortion identifier that predict the particular distortion and then estimate the quality score.[10].

      2. Histogram after normalization


    Non structural features like brightness, contrast and color accuracy is not noticed many of the times but by using the CIQE method it assess the Quality of the color in the image The contrast affect of an image is defined as


    The distance between the mass values of upper and lower parts of the luminance is presented on bin is the upper and lower mass values and is the average mass of luminance channel and brightness is measured according to the value

    of [1]

    Quality Score

    The color accuracy can be estimated by using the HSV model and the chromatic diversity is also noted. Then the spatial dispersion is obtained and the number of pixels having the dominant color is selected and from the distance between the pixels the spatial dispersion factor is obtained[13]. If the whole image have the dominant color then it has no chromatic diversity.


    By the combination of both structural and non stru.ctural distortion identification method is obtaine and measure the quality score.This method is called CIQE method Color image quality evaluation can be done by extracting the statistical features from the distorted images and then matches it with the existing value to obtain the quality score and the dispersion factor give the color accuracy.


    Input image

    Input image

    Extracting the computational features

    Extracting the computational features

    Proposed algorithm

    Proposed algorithm

    Image quality score computation

    Image quality score computation

    Quality measurement

    Quality measurement

    Compare the performance with other metric like MSE,PSNR,SSIM

    Compare the performance with other metric like MSE,PSNR,SSIM


    Fig.5. Flow chart for the Quality measurement algorithm.


    Step1.Selecting the input image.

    Step 2.Extracting the computational features.

    Step 3.Spatial dispersion of color image is also extracted.

    Step4.comparing it with the existing values

    Step 5.insert the signal to the SVM based distortion identifier.

    Step6.predict the type of distortion using quality predictor.

    Step 7.if the type of distortion is predicted then obtain the Quality score.


    From these structural and non structural distortions the overall quality is estimated and it want to be highly correlated with human perception.SVM and SVR is used to predict and classify the distortions. Hence the distortions like fast fading, jpeg compession,White noise Blurring affect and color accuracy of the image is identified and estimate the quality by SVM witha radial-basis function (RBF). Compare it with full reference method like PSNR and SSIM.The SSIM[1] shows Symmetric property as well as the boundedness property.

    1. Correlation affect of different images

    2. Comparison by PSNR

      The linear dependence measurement can be taken by the use of correlation coefficient in SSIM index

      .Overall correlation for the different images are then identified. The LIVE database is used to evaluate the performance of these no-reference metrics


    In this paper the CIQE method is used to distinguish the structural as well as the non structural distortion. The computational features are extracted for this and by spatial dispersion method the color accuracy is also identified. We classified basis on the structural features and then estimate the Quality score. This method is highly correlated with Human Perception. .The evaluation shows that the proposed no-reference metrics are robust in measuring the corresponding distortions


[1].zhou wang, member, ieee, alan c. bovik, fellow, IEEE Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, vol. 13, no. 4, april 2004.

[2] Z. Wang, A. C. Bovik, and L. Lu, Why is image quality assessment so difficult, in Proc. IEEE Int. Conf. Acoustics, Speech, and Signal Processing, vol. 4, Orlando, FL, May 2002, pp. 33133316

[3]. A DCT Statistics-Based Blind Image Quality IndexMichele A. Saad, Student Member, IEEE, Alan C. Bovik, Fellow, IEEE, and Christophe charrierieee signal processing letters, vol. 17, no. 6, june 2010.

[4].S. Suthaharan, No-reference visually significant blocking artifact metric for natural scene images, Signal Processing, vol. 89, no. 8, pp. 16471652, 2009

[5].Blind/referenceless image spatial quality evaluator anish mittal, anush k. moorthy and alan c.


[6].Saad, Michele A.Bovik, Alan C.;Charrier, Chris tophe Blind Image Quality Assessment: A Natural Scene Statistics Approach in the DCT DomainIEEE Transactions on Image Processing, vol. 21, issue 8, pp. 3339-3352

[7]. E. P. Simoncelli and B. A. Olshausen, Natural Image Statistics and Neural Representation, Annual Review of Neuroscience, vol. 24, no. 1,pp. 11931216, 2001.

[8]. R. Barland and A. Saadane, A reference free quality metric for compressed images, Proc. of 2nd Int. Workshop on Video Processing and Quality Metrics for Consumer Electronics, 2006.

  1. B. Schölkopf, A. J. Smola, R. C. Williamson, and P. L. Bartlett, New support vector

    algorithms, Neural Comput., vol. 12, no. 5, pp. 1207 1245, 2000.

  2. M. Narwaria and W. Lin, Objective image quality assessment based on support vector regression, IEEE Trans. Neural Newt., vol. 21, no. 3, pp. 515519, Mar. 2010

  3. MD Fairchild, A victory for equivalent background–on average. Proc Seventh Color Imaging Conference: Color Science, Systems and Applications (Scottsdale, 1999) 7, pp. 87

  4. E Peli, Contrast in complex images. J Opt Soc Am (10), 20322040 (1990)

  5. Sonia Ouni,Ezzeddine Zagrouba,Majed ChambahA New No-reference Method for Color Image Quality Assessment International Journal of Computer Applications (0975 8887) Volume 40 No.17, February 2012

Leave a Reply