978-1-4673-6469-0/13/$31.00 © 2013 243

No Reference Color Image Quality Measures Chen Gao, Karen Panetta, Fellow, IEEE

Electrical and Computer Engineering Department Tufts University

Medford, MA, 02155, USA Chen.Gao@tufts.edu, Karen@ece.tufts.edu

Sos Agaian, Senior Member, IEEE Electrical and Computer Engineering Department

University of Texas at San Antonio San Antonio, TX, 78249, USA

Sos.Agaian@utsa.edu

Abstract— Color image quality assessment is essential in evaluating the performance of color image enhancement and retrieval algorithms. Much effort has been made in recent years to develop objective image quality metrics that correlate with perceived quality measurements. Unfortunately, only limited success has been achieved [1]. In this paper we present: a) a new contrast based grayscale image quality measure: Root Mean Enhancement (RME); b) a color RME contrast measure CRME which explores the three dimensional contrast relationships of the RGB color channels; c) a color measure Color Qiality Enhancement (CQE) which is based on the linear combination of colorfulness, sharpness and contrast. Computer simulations show that the new measures may help to evaluate color image quality and choose the optimal operating parameters in color image processing systems. We demonstrate the effectiveness of the presented measures by using the TID2008 database. We also compare the presented measures with subjective evaluation Mean Opinion Score (MOS). Experimental results show good correlations between the presented measures and MOS.

Keywords-color measure, Root Mean Enhancement (RME), Color/Cube Root Mean Enhancement (CRME), Color Quality Enhancement (CQE)

I. INTRODUCTION Measuring the perceived image quality of a color image is

extremely difficult, because human vision is highly nonlinear and not equally sensitive to details in each color component. The most widely recognized method of determining color image quality is the subjective evaluation Mean Opinion Score (MOS). However, subjective evaluation is costly in time and resources thus it is difficult to be used in real time processing systems. The development of Computer Aided Design (CAD) made it desirable to formulate reliable objective color image quality measures. One would like to use such a measure so that it can evaluate the quality of an image and select optimal operating parameters automatically in color image processing algorithms.

Most existing color image quality measures are based on modifications of grayscale image quality measures. In general, grayscale image quality measures can be classified in three types: full reference approaches [2-6], reduced reference approaches [7] and no-reference approaches [8-16]. Full reference approaches require the original image as the reference and determines the similarity between the distorted image and the reference image. However, the cost of transmitting the original image is expensive, and the original images may not always be available. In the former situation, reduced reference measures can be used. The reduced

reference model reduces the original image to a limited number of parameters and transmits only the parameters [7]. In the latter situation, ideally, a 'no reference' measure is the best choice for image quality assessment. Most existing grayscale no reference measures are contrast based or sharpness based measures. Some methods use either Weber contrast [8] or Michelson contrast [9, 10];some measures are based on kurtosis [11, 12] or pixel derivatives [13]; some use the coefficients in DCT domain [14] or wavelet domain [15].

Some grayscale image quality measures have shown good correlation with human perceptions. So it is reasonable to utilize the grayscale image measurements such that these measures can be modified to measure the color image quality. Currently, the commonly adopted methods include transforming the color image to grayscale image and measuring the grayscale image quality or selecting a color space and linearly adding the measures on each color component. More sophisticated methods use different weights based on the content in each color component such as using PCA decomposition coefficients [17]. However, these simple conversions do not guarantee that the spatial correlation is preserved. While other methods may focus on one property of colorfulness, sharpness and contrast, this work strives to incorporate all three properties simultaneously in a measure. In this article, we propose two new methods: one combines the color metric, sharpness metric, and contrast metric together by multiple linear regression coefficients while the other treats the color image as a multi spectral signal and calculates the contrast or sharpness in a higher dimension.

The rest of this paper is organized as follows: Section II reviews existing grayscale and color image quality assessment methods. Section III proposes a new root mean square based contrast measure (RME) for both grayscale and color images and proposes a method to extend the 2D grayscale image measures to 3D color image measures. Section IV introduces another measure, Color Quality Enhancement (CQE) by linearly combining colorfulness, sharpness and contrast into single color image quality expression. Section V presents the experimental results and comparisons to the existing color measure Color Image Quality Index (CIQI) [18] and Mean Opinion Score (MOS). The conclusions are discussed in Section VI.

II. BACKGROUND Since grayscale image quality measures can be used and

extended to assess color image quality, A review of the grayscale image quality measures along with some existing color measures are provided in this section.

244 2013 IEEE International Conference on Cybernetics

A. Existing No Reference GrayScale Measures

In the past, no reference measures attempted to use statistical measures of the gray level distribution of local contrast enhancement. These include mean, variance, or entropy, but these approaches have not been particularly useful or meaningful [19]. Some more sophisticated no reference grayscale measures have been proposed [20-25]. These methods focus on specific types of distortion like blurring, JPEG compression or JPEG2000 compression. Morrow et al. [26] introduced a measure based on the contrast histogram, which has a much greater consistency than statistical measures. Following this observation, several measures of image enhancement have been developed by using a contrast measure.

The Weber contrast is shown in Eq.1 where �� represents the intensity change and � represents the local background intensity. It is used to measure the local contrast of a single target of uniform luminance seen against a uniform background [27]. The first practical measures which are based on the Weber’s law were the Measure of Enhancement (EME) and Measure of Enhancement by Entropy (EMEE). These measures were developed by Agaian [8, 28].

������ � (1)

Another commonly used contrast measure Michelson contrast is commonly used to measure the periodic pattern [27]. It is expressed in Eq.2 where ��� and ���� represent the maximum and minimum intensities in the whole image. DelMarco and Agaian created a visibility map by extending the Michelson operator to two dimensions [29]. Then the Michelson-Law measure of enhancement (AME), and Michelson-Law measure of enhancement by entropy (AMEE) [9] are proposed based on the Michelson contrast and logarithmic processing. Since Parametric logarithmic image processing (PLIP) [30] operations have been shown to be consistent with characteristics of human visual system and the nonlinear properties of image enhancement algorithms, Panetta et al. have developed the logarithmic Michelson contrast measure (logAME) and logarithmic AME by entropy (logAMEE) [10, 16, 31] by incorporating PLIP operations in AME and AMEE.

����������� �������������� (2)

To reduce the measure sensitivity to noise and steep edges, a second derivative based measure (SDME) was also introduced [13]. It measures the change ratio of the variation speed of pixel values and considers the center pixel value besides the local maximum and minimum value.

The definitions of these measures are listed in Table I. A general model is provided in [28]. Users can select the coefficients and functions for their applications. Each of the measures has different physical explanations within it, so it is necessary to choose a proper measure for a specific image type and application.

B. Existing No Reference Color Image Quality Measures

Much research has been conducted in recent years to develop objective color image quality metrics that correlate with perceived visual quality. Unfortunately, only limited success has been achieved [1]. Traditional methods of applying grayscale measures to color image measures is simply implemented by converting color image into grayscale image [3] or measuring the luminance component only. Some other methods measure the contrast in each color component individually and then combine the contrast measures of each color component by different weights [17]. However, these straightforward methods do not guarantee good performance.

Most previously proposed color image quality metrics focused only one aspect of image quality. Hasler and Susstrunk proposed a colorfulness measure in natural images by a linear combination of a subset of image statistics such as mean or standard deviation of chrominance and saturation [32]. Maalouf and Larabi proposed a color image sharpness metric in the wavelet domain which is tolerant to noise [15]. Bringier proposed a perceptual contrast assessment of color image based on the human visual system modeling [33].

Fu proposed a general color image quality index (CIQI) which is based upon linear combinations of three variables, colorfulness, sharpness and contrast [18]. Colorfulness is the attribute of a visual sensation according to which the perceived color of an image to be more or less chromatic. Fu's colorfulness metric is defined similarly as Hasler's [32] but normalized by a constant. Sharpness is defined as a function of the number of missed edge pixel between the estimated edge map and the real edge map. The more edge pixels missed, the smaller the sharpness metric value will be. Global contrast is modeled as the maximum local contrast and local contrast is defined as the ratio of the summation of high frequency band coefficients to the summation of low frequency band coefficients in the DCT domain. The coefficients of combining the three components are obtained by using multiple linear regressions. The training set includes 30 images with variety of distortions.

III. CUBE ROOT MEAN ENHANCEMENT (CRME) MEASURE In this section, we introduce a new enhancement measure

using the concept of Root Mean Square (RMS) contrast and a new methodology to extend the regular grayscale image enhancement measure to three dimensions so that it is applicable to measure color image quality.

1) New measures in grayscale image

The existing measures are either based on Weber contrast or Michelson contrast. Weber contrast is used to measure the local contrast of a single target of uniform luminance seen against a uniform background while Michelson contrast is commonly used to measure the contrast of period pattern [27].

2452013 IEEE International Conference on Cybernetics

TABLE I. EXISTING GRAYSCALE NO REFERENCE MEASURES

Grey Image Measure Definition Features

EME ���� �! "#$#%&&'()* +,�� -�-�,���-�-� .�!�/$

� �/$

Weber contrast based enhancement measure

Applicable to uniform background images [8]

EMEE ����0-� �! "#$#%&&1+,�� -�-�,���-�-� .0)* +,�� -�-�,���-�-� .�!�/$

� �/$

Weber contrast based enhancement measure

Incorporate entropy in EME More randomness in image, the bigger � to choose [8]

Visibility 2343536378 &&,�� -�-� 9 ,���-�-�,�� -�-� : ,���-�-��!�/$

� �/$

Two dimensional Michelson contrast based measure [29]

AME ;��� �! 9 "#$#%&&'()* +,�� -�-� 9 ,���-�-�,�� -�-� : ,���-�-�.�!�/$

� �/$

Michelson contrast based enhancement measure

Applicable to non-uniform images/ Applicable to Periodic background [9]

AMEE ;���0-� �! 9 "#$#%&&1+,�� -�-� 9 ,���-�-�,�� -�-� : ,���-�-�.0)* +,�� -�-� 9 ,���-�-�,�� -�-� : ,���-�-�.�!�/$

� �/$

Michelson contrast based enhancement measure Incorporate entropy in AME [9]

logAME 6<=;��� �! "#$#%>? && "'( )* +,�� -�-� @A ,���-�-�,�� -�-� BA ,���-�-�.�!�/$

� �/$

PLIP version of AME/ More appealing to human vision [10, 16]

logAMEE 6<=;���� �! "#$#%>? &&1+,�� -�-� @A ,���-�-�,�� -�-� BA ,���-�-�.0)* +,�� -�-� @A ,���-�-�,�� -�-� BA ,���-�-�.

�!�/$

� �/$

PLIP version of AMEE/ More appealing to human vision [10, 16]

SDME CD��� �! 9 "#$#%&&'()* EF ,�� -�-� 9 ',���G��-�-� : ,���-�-�,�� -�-� : ',���G��-�-� : ,���-�-�E�!�/$

� �/$

Second derivative based measure, It is not sensitive to noise [13]

HVS Visibility 2343536378HIJ K L���/$ >?M�+N�+�N� .N�+�N�!..�� BA 5�>?M�+N�+�N�O.N�+�N�P..��K QR�R/$ >?MR+NS+�NS .NS+�NS!..�S BA TR>?MR+NS+�NSO.NS+�NSP..�S A general visibility model [28]

However, in complex images these conditions are not always true. So we propose a new method based on the RMS which does not depend on spatial frequency content or distribution of contrast in the image.

Regular RMS contrast is a pixel-wised contrast as defined in Eq. 3. The new Root Mean Enhancement measure (RME) incorporates the idea of RMS contrast and the Human visual system (HVS) property. It measures the relative RMS contrast in the log domain (Eq. 4).

�U�J V $�W K K +,�-R 9 , X.%W�$R/Y��$�/Y (3)

Z�� $� �! [K K E��\]F ^�-S�_` a_`!aba_`�� ]��\]^�-S�_` a_`!aba_`�� ] E�!R/$� �/$ (4)

The image is divided into #$#% blocks, ,�-R is the center pixel intensity in block i, j, ^` �^`!�b�^`�� is the average intensity in block i, j. n is the total number of pixel within each block. It is worth noting that 6<= ]F ,�-R 9 ^` �^`!�b�^`�� ] returns a negative value if the difference between the center and its neighbors is smaller than 1. In the code implementation, we will use cdeF+"- ]F ,�-R 9 ^` �^`!�b�^`�� ]) to make sure the RME is always positive thus is proportional to the difference.

When the center pixel has an intensity value approximately equal to the average value of the current block, the contrast is small. So the RME measure is small for low contrast images and vice versa for images with high contrast. The recommended block size is 3x3 and no larger than 5x5 for the

reason that image statistical features are spatially nonstationary and image distortion is also space variant. When the RME is used for measuring image qualities, the bigger RME value indicates better contrast. When it is used to select parameters for computer aided algorithms, users are recommended to select the rightmost local peak. If multiple local peaks show similar RME values, an average should be taken among all candidates so it is called multi-scale RME (MSRME). This model is also quantitatively consistent with Weber’s Law, which states that the magnitude of a just-noticeable luminance change is approximately proportional to the background luminance for a wide range of luminance values.

2) New measure in color image

A color image is actually a multi spectral signal, and therefore it is reasonable to expand the traditional formulation of grayscale contrast measures so that it can be used to measure the multidimensional color image contrast. To be more specific, the contrast measures can be performed not only within each color plane but also across color planes. The cross plane contrast models the differences between color planes and reveals the color variation and structural difference raised from different color components. Based on this idea, we proposed the Color/Cube RME (CRME) to measure the relative difference of color cube center and all the neighbors in the current color cube:

�Z�� $YYY� �! [K K +��\]^�-S�K f`_` a_`!aba_`��Og ]��\]^�-S�K f`_` a_`!aba_`��Og ].0�!R/$� �/$ F (5)

246 2013 IEEE International Conference on Cybernetics

In the implementation, we use the intensity value of the center pixel as the cube center pixel intensity. h� represents the weights for different color components. In this article we use Fh� (i'jj , Fh\ (iklm and h� (i""n in accordance with NTSC standard. According to the Human Visual System (HVS) properties [34], human eyes are not equally sensitive to the just noticeable difference (JND). So an � power is assigned to the ratio based on the region to which the local background of the current cube belongs. If it belongs to the DeVries-Rose region, �=0.2; if it belongs to the Weber region, �=0.4; and if the background intensity falls into the Saturation region, �=0.8.

IV. COLOR QUALITY ENHANCEMENT (CQE) MEASURE Next, we introduce another color image quality measure

CQE. CQE uses a polling method to combine chrominance information with measures of sharpness and contrast. The idea is similar to [18] but differs in many ways. We implemented the algorithm described in [18] and observed that their attribute metrics do not have strong correlations with human visual perceptions. Thus we use different colorfulness, sharpness and contrast metrics in the CQE measure (Eq. 6).

�o� Q$ p Q<6<qrs6tu44 : Q% p 4vLqwtu44 : Qx p Q<t7qL47 (6)

Colorfulness is the attribute of chrominance information humans perceive. Hasler and Susstrunk have shown that the colorfulness can be represented effectively with combinations of image statistics [32]. Our colorfulness measure is defined in Eq. 7. It is formulated as the ratio of the variance to the average chrominance in each of the opponent component.

Q<6<qrs6tu44 (i(' p )yzF+ {1'|}1|(i'. p )yzF+ {~']}~](i'.

( 7 )

where 1 Z 9 M- ~ (ik+Z : M. 9 � are opponent spaces and , {0% ,F{�% , �0 , �� represent the variance and mean values along these two opponent color axes.

Sharpness is the attribute related to the preservation of fine details. It is related with acutance and resolution in each color component. In our method, we first apply the Sobel edge detection algorithm on each RGB color component and then multiply the binary edge map with the original values to obtain three grayscale edge maps. Since sharpness is proportional to the perceived steepness of slopes, it is reasonable to treat each of the grayscale edge maps as an input image and measure the Weber contrast if the window size is small enough. In this article, the EME measure is used on each grayscale edge map with an overlapped window size 3x3. It is worth noticing that if the input image is a JPEG compressed image, the window which is overlapped with JPEG artifact block borders will be discarded. In Eq.8, the sameFh� coefficents as in CRME are used.

4vLqwtu44 K h�������������+=qL8uT=u�.x�/$ (8)

������������ %� �! K K 6<=F+^���-�-�^���-�-� .�!�/$� �/$ (9) Contrast is usually defined as the ratio of maximum

intensity to the minimum intensity of the entire image or local

regions. We measure the contrast by applying the AME measure in the intensity component. The original AME expression returns smaller values for images with larger contrast. Therefore, we add a power of -0.5 to each term so that the better contrast image has larger contrast metric value.

Q<t7qL47 ;�����G���G+,t7ut4378. (10)

;�����G���G $� �! K K +6<= �^���-�-��^���-�-�^���-�-��^���-�-��.�Yi��!�/$� �/$ (1 1 ) After the colorfulness, sharpness and contrast metrics are

obtained, we then apply a multiple linear regression (MLR) on 48 samples of training images from TID2008 database to obtain the three coefficients in Eq. 6. For an image with a specific type of distortion, the CQE measure achieves better performance if the training set has the same distortion. For images with unknown distortions, a more general set of coefficients can be used. The recommended coefficients for some types of distortion and for generic usage are listed in Table. II.

TABLE II. RECOMMENDED COEFFICIENTS FOR APPLYING CQE MEASURES ON DIFFERENT TYPE OF DISTORTED IMAGES. THE COEFFICIENTS

ARE OBTAINED BY APPLYING MLR ON 48 TRAINING IMAGES SUFFERING FROM THE CORRESPONDNG DISTORTIONS

Contrast Change Gaussian Blur JPEG2000

Compression Generic

Distortions

c1 0.4358 0.2736 0.2170 0.2946 c2 0.1722 0.2261 0.7100 0.3483

c3 0.3920 0.5003 0.0731 0.3571

V. EXPERIMENTAL RESULTS The TID2008 database [35, 36] is a published benchmark

for measuring color image qualities and it is used in this paper to test the effectiveness of the color measures. It is worth noting that since all measures have different dynamic ranges, we normalize all measure values for comparison.

Figure 1. measure values for 48 Gaussian blurred images. Blue: MOS score; Red: CRME score; Green: CIQI score; Megenta: CQE score

2472013 IEEE International Conference on Cybernetics

Figure 2.Comparisons of measure values for images suffering from different levels of distortions. (a)-(d) shows four images with different levels of distortions and their measure values; (e)demonstrates the normalized measure values for the four images. First row are contrast changed images, second row are JPEG2000

compressed images, third row are Gaussian blurred images

TABLE III. COMAPRISON OF MEASURE MATRICS IN ACCORDANCE WITH AVERAGE PEARSON, SPEARMAN AND KENDALL CORRELATION WITH MOS ON

THREE TYPES OF DISTORTIONS: GAUSSIAN BLUR, CONTRAST CHANGE, JPEG2000 COMPRESSION IN TID2008 DATABASE

Distortion Type Measure Pearson

Correlation Spearman

Correlation Kendall

Correlation

Gaussian Blur

CRME 0.9882 1.0000 1.0000

CIQI 0.6527 0.5833 0.5278

CQE 0.9631 1.0000 1.0000

Contrast Change

CRME 0.9569 0.8333 0.7500

CIQI 0.9099 0.8167 0.7222

CQE 0.9164 0.8333 0.7500

JPEG2000 compression

CRME 0.9891 1.0000 1.0000

CIQI 0.6945 0.7667 0.6944

CQE 0.9255 0.9833 0.9722

CRME shows good performance in evaluating an image quality if the image is distorted by low frequency factors. High frequency distortions such impulse noise will introduce high frequency components thus the CRME measure is not applicable. Fig. 1 shows the measure values of MOS, CRME, CIQI and CQE for 48 Gaussian blurred images in TID2008 database. Table III demonstrates the average Pearson correlation, Spearman correlation and Kendall correlation for 48 Gaussian blurred images, 48 contrast change images and 48 JPEG2000 compressed images. The results show that the proposed two measures have good correlation with the MOS and perform better than CIQI.

The color image quality measures can also be used to rank image qualities. Some images with different levels of distortions (Contrast change, JPEG2000 compression and Gaussian blur from top row to bottom row) and the corresponding measure values are shown in Fig.2. In this point

MOS = 6.5000

CRME = 1.3873

CIQI = 0.5327

CQE = 1.2122

MOS = 4.9714

CRME = 1.2716

CIQI = 0.4338

CQE = 0.7680

MOS = 6.1944

CRME = 1.4268

CIQI = 0.6026

CQE = 1.4771

MOS = 3.9722

CRME = 1.1543

CIQI = 0.3930

CQE = 0.5545

MOS = 5.0625

CRME = 0.7371

CIQI = 0.4080

CQE = 0.8582

MOS = 2.9355

CRME = 0.4249

CIQI = 0.3437

CQE = 0.7163

MOS = 1.2333

CRME = 0.1450

CIQI = 0.3370

CQE = 0.5014

MOS = 0.0000

CRME = 0.0078

CIQI = 0.3210

CQE = 0.3614

MOS = 5.7931

CRME = 0.8570

CIQI = 0.5666

CQE = 0.8672

MOS = 5.2414

CRME = 0.7077

CIQI = 0.5914

CQE = 0.6561

MOS = 4.0000

CRME = 0.5073

CIQI = 0.5902

CQE = 0.4631

MOS = 3.0690

CRME = 0.2781

CIQI = 0.5144

CQE = 0.3704

(a) (b) (c) (d) (e)

248 2013 IEEE International Conference on Cybernetics

of view, the measures can also be used for assist selecting optimal parameters in image enhancement, denoising or decompression algorithms.

VI. CONCLUSION In this article we reviewed existing image quality measures

and presented two new color measures: CRME and CQE. We compared the CRME and CQE with CIQM and MOS on TID2008 image quality database. Experiment results showed that the new measures have good correlation with MOS, and they can be used to select optimal parameters for image processing algorithms. Both measures are reliable to assist computer design. Also, while the CRME measure has fast processing time so that it can be used in real time image processing systems, the CQE measure is applicable to a wider variety of images.

ACKNOWLEDGMENT This work has been partially supported by NSF Grant

HRD-0932339. REFERENCES

[1] Z. Wang, et al., "Why is image quality assessment so difficult?," in Acoustics, Speech, and Signal Processing (ICASSP), 2002 IEEE International Conference on, 2002, pp. IV-3313-IV-3316.

[2] D. Chandler and S. Hemami, "VSNR: A Wavelet-Based Visual Signal-to-Noise Ratio for Natural Images," IEEE Transactions on Image Processing, vol. 16, September, 2007.

[3] Z. Wang, et al., "Image Quality Assessment: From Error Visibility to Structural Similarity," IEEE TRANSACTIONS ON Image Processing, vol. 13, April, 2004.

[4] E. A. Silva, et al., "Quantifying image similarity using measure of enhancement by entropy," Mobile Multimedia/Image Processing for Military and Security Applications, vol. 6579, p. 65790U, 2007.

[5] S. Nercessian, et al., "An Image Similarity Measure Using Enhanced Human Visual System Characteristics," in Mobile Multimedia/Image Processing, Security, and Applications, 2011.

[6] Z. Wang and A. C. Bovik, "Mean squared error: love it or leave it? A new look at signal fidelity measures," Signal Processing Magazine, IEEE, vol. 26, pp. 98-117, 2009.

[7] Q. Li and Z. Wang, "General-purpose reduced-reference image quality assessment based on perceptually and statistically motivated image representation," in Image Processing, 2008. ICIP 2008. 15th IEEE International Conference on, 2008, pp. 1192-1195.

[8] S. S. Agaian, et al., "A new measure of image enhancement," presented at the IASTED International Conference on Signal Processing & Communication, 2000.

[9] S. S. Agaian, et al., "Transform coefficient histogram-based image enhancement algorithms using contrast entropy," Image Processing, IEEE Transactions on, vol. 16, pp. 741-758, 2007.

[10] K. Panetta, et al., "Human visual system-based image enhancement and logarithmic contrast measure," IEEE Transactions on Systems, Man, Cybernetics, vol. 38, Feb. 2008.

[11] J. Caviedes and F. Oberti, "A new sharpness metric based on local kurtosis, edge and energy information," Signal Processing: Image Communication, vol. 19, pp. 147-161, 2004.

[12] J. Caviedes and S. Gurbuz, "No-reference sharpness metric based on local edge kurtosis," in Image Processing. 2002. Proceedings. 2002 International Conference on, 2002, pp. III-53-III-56 vol. 3.

[13] Y. Zhou, et al., "Nonlinear unsharp masking for mammogram enhancement," IEEE Transactions on Information Technology in Biomedicine, vol. 15, pp. 918-928, November, 2011.

[14] J. Tang, et al., "Image enhancement using a contrast measure in the compressed domain," Signal Processing Letters, IEEE, vol. 10, pp. 289-292, 2003.

[15] A. Maalouf and M.-C. Larabi, "A no reference objective color image sharpness metric," in EUSIPCO, 2010, pp. 1019-1022.

[16] E. Wharton, et al., "A logarithmic measure of image enhancement," in Mobile Multimedia/Image Processing for Military and Security Applications, 2006.

[17] C. Gao, et al., "A new color contrast enhancement algorithm for robotic applications," presented at the IEEE conference on Technologies for Practical Robot Applications, 2012.

[18] Y.-Y. Fu, "Color image quality measures and retrieval," Ph.D, Department of Computer Science, New Jersey Institute of Technology, January 2006.

[19] T.-L. Chen, "On the statistics of natural images," Brown University, 2005.

[20] P. Marziliano, et al., "A no-reference perceptual blur metric," in Image Processing. 2002. Proceedings. 2002 International Conference on, 2002, pp. III-57-III-60 vol. 3.

[21] D. S. Turaga, et al., "No reference PSNR estimation for compressed pictures," Signal Processing: Image Communication, vol. 19, pp. 173-184, 2004.

[22] Z. Wang, et al., "No-reference perceptual quality assessment of JPEG compressed images," in Image Processing. 2002. Proceedings. 2002 International Conference on, 2002, pp. I-477-I-480 vol. 1.

[23] R. Sakuldee, et al., "Image quality assessment for JPEG and JPEG2000," in Convergence and Hybrid Information Technology, 2008. ICCIT'08. Third International Conference on, 2008, pp. 320-325.

[24] H. Tong, et al., "No-reference quality assessment for JPEG2000 compressed images," in Image Processing, 2004. ICIP'04. 2004 International Conference on, 2004, pp. 3539-3542.

[25] A.Grigoyan and S.Agaian, "Image Enhancement," in Advances in Imaging and Electron Physic, ed: Academic Press, May 2004, pp. 165-243.

[26] W. M. Morrow, et al., "Region-based contrast enhancement of mammograms," IEEE transactions on Medical Imaging vol. 11, pp. 392-406, Sep. 1992.

[27] E. Peli, "Contrast in complex images," Journal of Optical Socience of America, vol. 7, October 1990.

[28] S. S. Agaian, "Visual morphology," in Electronic Imaging'99, 1999, pp. 139-150.

[29] S. DelMarco and S. Agaian, "The design of wavelets for image enhancement and target detection," Mobile Multimedia/Image Processing, Security, and Applications, pp. 735103-12, 2009.

[30] K. Panetta, et al., "Parameterized logarithmic framework for image enhancement," IEEE Transactions on Systems, Man, and Cybernetics—Part B: Cybernetics, vol. 41, April 2011.

[31] E. Wharton, et al., "Comparative study of logarithmic enhancement algorithms with performance measure," in Electronic Imaging 2006, 2006, pp. 606412-606412-12.

[32] D. Hasler and S. E. Suesstrunk, "Measuring colorfulness in natural images," in Electronic Imaging 2003, 2003, pp. 87-95.

[33] B. Bringier, et al., "No-reference perceptual quality assessment of colour image," in Proceedings of the European Signal Processing Conference (EUSIPCO'06), 2006.

[34] G. Buchsbaum, "An analytical derivation of visual nonlinearity," IEEE Transactions on Biomedical Imaging, vol. BME-27, pp. 237-242, May 1980.

[35] N. Ponomarenko, et al., "Metrics performance comparison for color image database," in Fourth international workshop on video processing and quality metrics for consumer electronics, 2009.

[36] N. Ponomarenko, et al., "Color image database for evaluation of image quality metrics," in Multimedia Signal Processing, 2008 IEEE 10th Workshop on, 2008, pp. 403-408.