水下图像评价指标（PDF文件）

【实例简介】
UCIQE方法一致认为具有更好的清晰度、对比度和亮度，更加丰富的色彩和自然的表面，并且远景目标得到更好揭示的图像具有更高的质量分数，这一点符合人类的视觉感知。
6064 IEEE TRANSACTIONS ON IMAGE PROCESSING. VOL, 24. NO. 12 DECEMBER 2015 including absorption and scattering by phytoplankton absorption by coloured dissolved organic matter(CDOM) and finally, light scattering by total suspended matter (TSM)[32] Forward scattering (randomly deviated light on its way from an object to the camera) generally leads to blurring of the image features. On the other hand, backward scattering(the fraction of the light reflected by the water towards the camera before it actually reaches the objects in the scene) generally limits the image contrast, generating a characteristic veil that superimposes itself on the image and hides the scene Floating particles(marine snow) increase the absorption and sCattering effects. As a result of different absorption spectra, the reflection of colours will vary between different water types depending on the contribution from the different Inside Optical Parameters(IOP). The concentration of IOP and the distance to the object of interest are therefore important hen evaluati ality [33]. The visibility range can be increased with artificial lighting but these sources not only suffer from some scattering and absorption, but in addition tend to illuminate the scene in a non-uniform fashion producing bright spots in the image and poorly illuminated areas surrounding the spots. As depth increases, colours drop off one by one depending on their wavelengths. First, red colour disappears at a depth of 3m approximatelY. At 5m, Fig Underwater colour images, and their hue histograms in polar orange colour is lost. Most of the yellow goes off at 10m and finally the green and purple disappear at further depth Blue colour travels the longest in water due to its shorter where wavelength. Underwater images are therefore dominated by blue-green colour. Also the light source variations will affect C!@l_colorfulness -(od+B+0.3u&+up)/85.59 colour perception. As a consequence, a strong and non uniform colour cast characterizes the typical colour distortion cIgI_sharpness of underwater images [4]. Finally, the underwater engineering =l-(1-(tepestimated-tepsobel/tepsobei 0.2 and monitoring colour images are chroma decreased and hue (5) shifted towards blueness, non-uniform cast. blurring and noise. ClQlcontrusl a group of typical underwater monitoring and survey images and their polar hue histograms are shown in Fig. 1. It max( ocal_contrast∑Bond/∑Bond) be seen that the distributions of hues are non-uniform prominently blue-green or yellow and 方,μa, represent the variance and mean values along the two opponent colour axes defined in(1) and(2) B. Colour Image qualily Metrics /or Aimospheric images tepestimated denotes number of edge pixels estimated; tepsobel denotes number of edge pixels counted using Sobel operator Hasler and suesstrunk [16] show that colourfulness can be Bond; is the i th coefficient of the total 15 bands of 8x 8 blocks represented effectively with a combination of image statistics. of DCT coefficients. c1, C2, c3 are weighted coefficients This feature is incorporated to our new metric. Fu [17] and CQE metric [18 is similar to the CIQI measure but differs Panetta et al. [18] define colourfulness in the opponent colour in the colourfulness, sharpness and contrast definitions space with red-green channel and yellow-blue channel. For an RGB image 1, let a denote the rg channel, and b denote the CQE=c1× COE_colorful nes s yb channel x CE_ sharpness +C3 X COE_contrast (7) a=R-G (1)where B=0.5×(R+G)-B (2) CQE_colorfulness Based on the opponent colour space, Fu [171 combined =0.02×log chrominance information with sharpness and contrast and uB proposed the ciQ metric defined b C1Ql=C1 x ClQl_color fulness +c COE-sharpnes s=>heEM Sharpness(grayedgec x CIQI_Sharpness +c3 C1QI_contrast (3) YANG AND SOWMYA: UCIQE METRIC 065 TABLE the different distortion levels for similar image content;(e)has PERFORMANCE OF EXISTING COLOUR IMAGE QUALITY METRICS low computational complexity and can be implemented in real Index time 6 Metrics A. The colour statistics metrics -0.28530.0040.007200159042130.2311 0.44090.0162007670.00150.24180.2139 CIELab space is a uniform colour space and device inde 0.0015 0.0336 0.00031 0.0003 0.0046 0.0007 pendent. Hasler and Suesstrunk [16] studied twelve metr 0.0039 0.0154 0.0060 0.0006 0.0033 0.0011 of image pixels in the CIELab colour space, including the CQE0.08290.11040.02710.00530.03000.0607 CRME5.04269.64288.76276.32133.155592603 standard deviation along the a axis, b axis, chroma and CIQI 0.0672 0.2037 0.0693 0.0119 0.0877 0.0388 saturation, the mean of chroma and saturation and so on Since they assume that image colourfulness can be represented by a linear combination of a subset of these metrics, to k∑∑ maxk l EME Sharpness log( (10) find the best correlated metric for degradations in underwater minkl monitoring and survey colour images, a set of subjective tests C OE_contrast AMEcontrast(intensity) (11) were conducted as follows 44 underwater images were shown to human observers AME_ contrast=∑>og max, k, /+Imin, k, I-0.5 These images were obtained from different underwater maxk.1-lmink. environments including pipeline detection in muddy water, (12) d shallo micro particles is usually high. The lighting conditions include natural kI x k2 is the size of the image block, and Ik. is the pixel day lighting, lighting with green laser and LEd white-light intensity in the image block. ic represents the weight for sources. The contents are varied. The distortions include different colour components. Panetta et al. [18] also expand blurring, low contrast, low saturation, colour cast, marine the grayscale contrast measures to the multidimensional colour snow and motion muddy caused by underwater creatures image contrast and propose the crme to measure the relative Some of these images are shown in Fig. 1 difference of the colour cube centre and all the neighbours in The images were randomly displayed; for each displayed the current colour cube The crme metric is mage, the subject was asked to rate the image quality using a scale from 1 to 5 corresponding to "Very annoying, 100|g1,-∑A++L “ Annoying,”“ Slightly annoying,”“ Perceptible but not CRME= ∑∑ annoying, and"Imperceptible,, respectively. In order to =lj=110g ;i+)21c1+12++I reduce the effect of outliers, each image was presented 4 times. A subject could not proceed to the next image until e(13) the current image was scored 12 subjects took the test and the Mean Opinion Score(mos) was computed where, Ii. i is the centre pixel intensity in the block and n is For each subjective level,9 CIELab space statistics were the total number of pixels within each block computed including average of chroma uc, variance of In marine habitats, the rough absorption of the colours chroma oc, average of saturation us, variance of saturation as toward the red end of the spectrum lowers the value of the red a pseudo-area in ab space, the standard deviation along the a component in RGB space as the depth increases. For all these and b axis, the root Mean Enhancement(RME)contrast [18] three colour image metrics, as the red component decreases in of a and b, and the contrast of l channel, as shown in Fig. 2 underwater images, the value of a panel will be negative and The histograms of these metrics larger than the average are the absolute value will increase. Marine snow with artificial shown in Fig. 2(a). It can be seen that, for the underwater lighting will cause increased local contrast and a wrongly high monitoring and survey images, the mos are generally low quality value. The statistical values of a, B and performance For images with higher MOs, the oc, contrast of l, O, and us results of CQE, CRME and CIQI metrics for the images shown are all higher than the averages, and they change linearly in Fig. 1 are in Table 1. The data reveals that these natural with decreasing mos In Fig. 2(b), the mean values of these colour image quality metrics fail to predict the degradation of g metrics with different MOs levels are shown. Clearly the underwater images. For example, all of these three metrics that the mean values of oc, the contrast of l,os and u give a higher score to the 6th image, while only snowing noise increase linearly with the Mos. That is to say, for underwater can be seen in it monitoring and survey colour images the deviation of hue the contrast of brightness and saturation correlate well with the IV. UNDERWATER COLOUR IMAGE observers' perceptions. In addition, the statistics that correlate QUALITY EVALUATION METRIC with Mos will change with different environments and degra- One would like to use a measure in the underwater moni dation features. For example, colourful seafloor photography toring and survey colour image analysis that:(a) is correlated images(not included in the 44 images) have generally higher with human perception;(b) is suitable for classical types of MOS than others, as variance of saturation Os, sharpness of distortion of images taken in turbid water;(c) is reliable for luminance and a, b channel, mainly determine the extent of underwater image enhancement processing;(d) can measure obServers visual perception. 6066 IEEE TRANSACTIONS ON IMAGE PROCESSING. VOL, 24. NO. 12 DECEMBER 2015 space is defined as erage of chroma ME of a UCIQE=C1Xc+c2×conl+c3×p 0.8 RME of b vanance of saturatio where, oc is the standard deviation of chroma, con is the brigh. ness contrast contrast of luminance and us is the average of saturation 0.6 variance of chroma and c1, C2, c3 are weighted coefficients. As described above aerage of saturation the variance of chroma has good correlation with human perception for underwater colour images of interest. There are also other reasons for adopting the variance of chroma to describe the colour cast. one reason is that for colour images taken in muddy water with artificial lighting, marine snow is notably a major source of image degradation as the scattering creates white bright spots that may strongly impact the performance of image processing methods. The common Image subjec: ive leve metrics based on contrast and gradient will give higher scores However, the hue distribution will not be influenced by marine snow. Tank images taken in 680cm transparencies of water with increasing camera distances are shown in Fig 3(a). The corresponding hue channels are shown in Fig 3(b) and the +… ave ace of chroma histograms of hue can be seen in Fig 3(c). The data shown in Fig. 3(c)show that with increased camera distances, the 05|-△- nghtness contrast variance of hue decreases, al though there are more spots in …aran the image with increased camera distance Contrast is used to measure the local contrast of a single target seen against a uniform background. It is one of the most perceived factors when the water environment is mudd and particle rich. Here, con! was computed by the difference between the bottom 1% and the top 1% of all pixel values in luminance channel. The value returned can represent the Image subjective level global gray distribution of an image After the standard deviation of chroma, contrast and average Fig.2.The distribution of CIELab space statistics against the Mos. of saturation are obtained, for the 44 test image data se (a) histograms of nine metrics higher than the average values of different 4-fold cross-validation was performed, three folds were used MOS level groups; (b)average values of nine metrics for different mos level for training and a multiple linear regression(MLr) on training group images from subjective data was applied to obtain the three coefficients(14). The last fold was used for evaluation. This B. Proposed Underwater Colour Image process was repeated 4 times, leaving a different fold for Quality Evaluation metric evaluation each time and the median of the values across iterations is reported. It is observed that the contrast. chroma In this work, the underwater colour images of concern and saturation are calculated independently so they can be are the raw images taken in underwater pipeline monitoring processed in parallel to accelerate computation speed. For or engineering survey. Most of these underwater images are underwater monitoring and survey colour images with blur blurred, have low contrast and severe colour cast. To select ring, colour cast and marine snow distortions, the obtained the best metric, several aspects must be considered: the coefficients are c1=0.4680, C2=0. 2745, C3=0.2576. For other most obvious is the correlation to the subjective test data, underwater colour images with a specific type of distortion the second is the computational cost, and the last is related the UCIQE with different metrics combination achieves better to the limitation of the experiment due to the initial choice performance if the training set has the same distortion in the selection of the 44 scenes. As the CIELab space is To obtain the performance of the major natural colour image designed to be a uniform colour space, it does not seem quality metrics on underwater images, MLR was also applied reasonable to emphasize the blue-yellow axis, as described by on the 44 testing underwater images to get the optimized Hasler and Suesstrunk [161 coefficients in(3)and(7). For CiQI metric, C1=0, C2=0.5141 It is also reasonable to avoid using deviation of satura- C3=0.4859. For CQE metric, c1=0, C2=0.2351, c3=0.7649 tion as, since it over-emphasizes dark areas, which are pre- The optimization results also indicate that for images taken in cisely the areas that some underwater images contain because turbid water with high concentration suspended matter, sharp of limited lighting. Let Ip be the pixel values of an image ness and contrast are more important than the colorfulness in CIELab space, p= 1. N. The image has N pixels. for perceptual image quality in CIQI and CQE metrics.The Ip=llp, ap, bp]. CI is the chroma [16]. The underwater colour experimental results of CIQI and CQE listed in the next section image quality evaluation metric UCIQE for image I in CIELab are computed with the optimized coefficients YANG AND SOWMYA: UCIQE METRIC 067 variance of hue is 0 070 ariance of hue is 0. 0673 vanance of huc s 0. 053 variarce of hue is 0.0614 0,6 268号 号号 D。gre 330clll and 360cIn far froln camera, respectively. (b)The corresponding hue channels. (c)The histograMs \S wr oard images with 240cm, 270cm, 300cm, including CIQI, CQE and CRME. The proposed UCIQE o13元 is also compared with WGSA [26 and gradient magnitude histogram metric(R)[27, although they were designed for 24丁 grayscale underwater image restoration. Part of the testing board and colour chart images taken in clear and medium Fig. 4. Tank and targets muddy water with increased camera distances are shown in Figs. 5 and 7. Corresponding values of image quality with increased camera distances for the two sequences are plotted V RESULTS AND DISCUSSION in Figs. 6 and 8. With the increased camera distances, the The experiments were divided into three parts. The first attenuations are more serious, and the added artificial lighting aggravated the back scattering and noise degradation as shown series of experiments were conducted to confirm the accuracy of the proposed metric for predicting different degradations in Figs. 5 and 7. While the natural colour image quality metrics mentioned in this paper failed to predict the degradation with increased camera distances. The second part is subjective experiments, to compare the perceptual relevance of the met tendency as shown in Figs 6 and 8. For example, non-uniform light spot and the strong backscattering of suspended matter ric. The third part is to evaluate the suitability of the proposed as shown in Fig. 5(c) increases the contrast value in CiQI metric for underwater image enhancement algorithms measure and result in a deviating point as pointed in Fig 6.(c) The curves plotted in Figs. 6(f) and 8(f)illustrate that A. Tank tests the proposed metric UCIQE indicates the linear change more The tank is 2.53m long, 1.02m wide, 1.03m high, with two accurately than the three leading colour and the two grayscale observation windows measuring 33cm diameter on both sides underwater image quality metrics. Whereas, note that WGsA of the tank. The images were taken with OTI-UWC-325/P/E andR were applied to test images after transforming colour colour camera, and the artificial lighting source is a 500w images to grayscale images first halogen lamp. Several sequences of images(960x 576)were taken under different conditions, including 680cm. 190cm and B. Subjective experiments 94.5cm transparencies of water [ 34, in natural and artifi cial lighting with board and Color Checker 24 X-Rite Chart To get meaningful results, it is important not to use the (21.59 x27.94cm)targets, as shown in Fig 4 same data in computing the correlation and in optimizing An attempt was made to compare the proposed metric the parameter set. As mentioned above, when applying 4-fold UCIQE to other state-of-art colour image quality metrics cross-validation, one of the four folds is used to compute the 6068 IEEE TRANSACTIONS ON IMAGE PROCESSING. VOL, 24. NO. 12 DECEMBER 2015 2 3T 4T (a) 论四 water with led lighting.(a)90cm(b) 120cm(c)150cm(d)180cm.y of Fig. 5. Samplcs of board imagc scqucncc takcn in 680cm transpar Fig. 7. Color Chcckcr chart images taken in 190cm transparency of watcr with led lighting.(a) 90cm.(b)120cm(c)150cm(d)180cm Camera Do: need<m) Cimera distance/cm) Carrea Distante/cm (b) 忑5b Camera Distancacm Fig. 6. Quality values of board images(Fig. 5).(a)R ( b) WGSA(c)CRMe (d) CQE.(e)CiQI (f)UCIQE Fig. 8. Quality values of ColorChecker chart images (Fig.7).(a)R (b)WGSA.(c)CRME.(d CQE.(e)CIQI (f)UCIQE correlation of the metrics with the experiment data The objec- tive quality predictions do not map directly to the subjective mean opinion scores (MOs)and there is a non-linear mapping function between subjective and objective predictions. A cubic the accuracy of the image qualities [35]. The results are sum polynomial with four parameters is fitted to account for marized in Table II in terms of PRCC, rmse and SRCC. The this mapping. Common correlation coefficients are used to results show the superiority of the proposed uCiQe metric in analyse the statistical relationship between two sets of images. terms of accuracy, monotonicity and consistency, as compared Pearsons product moment correlation(PRCC) measures how to the existing metrics for underwater pipeline monitoring and far each measure value deviates from the MOs. Spearman,s survey colour images. The proposed metric has good correla- rank order correlation(SRCC) compares the rank of image tion with Mos on the order of 0.76 and performance range qualities and the root mean square error(RMSE)measures from 20, 9 and 25 percent better than CQE, CRME and CiQI YANG AND SOWMYA: UCIQE METRIC 069 (a) (d) Fig 9. Real underwater monitoring images enhancement test. (a) Raw images(b) He et al.(c) Fattal et al.(d) Tarel et al.(e) Iqbal et al.(f)Ancuti et al TABLE II TABLE III PERFORMANCE COMPARISON OF PROPOSED METRIC COMPARISON OF ENHANCEMENT METHODS EVALUATION ON REAL WITH STATE-OF-ART COLOUR METRICS UNDERWATER IMAGES PRCC R SRCC Target Metric Image 1 Image 2 Image4 UCIQE 0.7549 0.0837 0.7543 0.3048 0.30640.0677 0.1293 CQE 0.557 0.1027 0.5331 Degraded CRME 3.1555302533.0457 CRME 0.6945 0.0917 0.5823 Image -0.0465-0.0285 0.0427 0.0377 CIQ 0.5626 0.1187 0.2832 UCIQH0.540004775039300.5009 CIOI 0.3060 0.3078 0.0935 0.1544 He CRME 3.0217 3.1855 3.124 3.1131 [36] CQE -0.0421-0.02480.0390-0.0346 UCIQE0.58740.5228 04818 0.5631 C Image enhancement Results evaluation 0.3106 0.3139 0.1055 0.2031 There have been lots of attempts to enhance the visibility of Fattal CRME 2.9527 3.1783 2.8858 3.0473 37」 -0.04180.0220-0.0274-0.0211 single degraded underwater colour images, such as defogging UCIQE 0.6448 0.5439 0.6158 0.6503 based algorithms 361, 371, contrast stretching meth CIQI 0.3087 0.3090 0.08380.1288 ods [381, [39 and the newest image fusion enhancement [6] Tarel CRME 3. 0489 3.2361 3.1884 3.1535 The capability of the proposed UCiQE as an effective metric 38 CQE -0.06900.04220.0538 0.0438 to measure the image enhancement results was tested Five UCIQE0.58210.50460.53080.5828 IQI 0.5177 0 0.5035 0.4674 underwater image enhancement algorithms were presented Iqbal CRME 2.8687 2.7290 2.6302 2.7417 including: scene depth information-based dark channel prior [39 COE -0.0403002870.07340.0405 dehazing method proposed by He et al. [36], single image UCIQE 0.7684 0.6797 0.5919 0.7507 dehazing algorithm proposed by Fattal [37, fast visibility CIQ 0.4061 0.3755 0.488806216 Ancuti CRMe 2.8670 2.8390 2.6644 2.7973 restoration method proposed by Tarel and Hautiere [38 and -0.0280 0.02720.03450.0465 underwater colour image enhancement method based on UCIQE 0.8937 0.8551 0.7441 0.8814 integrated model proposed by lqbal et al. [39 and the fusion based method [6]. a group of underwater degraded images and corresponding enhancement processing results are shown Table IIl. The data verifies the better coherence of UCIQE in Fig 9. Among those enhancements methods compared, with the subjective perspective than the others the images enhanced by image fusion method [6] obtain In Table IV. the average execution time for 60 underwater comparably better results. Comparisons of different colour colour test images is shown. The size of the test images image quality evaluation approaches with UCIQE are list in is 960 x576x3, tests are on 2.8 GHz frequency Intel i7 6070 IEEE TRANSACTIONS ON IMAGE PROCESSING. VOL, 24. NO. 12 DECEMBER 2015 TABLE IV [9] Y. Wang, Q. Chell, and B. Zhang, Ilage enhanIceinent based on equal AVERAGE EXECUTION TIME FOR THE UCIQE area dualistic sub-image histogram equalization method, IEEE Tran. CQE, CRME AND CIQI Consum. Electron., vol. 45, no. 1, pp. 68-75, Feb. 1999 [10] M. Kim and M. Chung. Recursively separated and weighted his- UCIQE C妮 E CRME CIQI lograIn equalization Tor brightness preservation and contrast enhance Average execution(s) 0.20 7.09 8.65 0.83 ment, IEEE Trans. Consum. Electron., vol 54, no 3, Pp. 1389-1397 Aug.2008 [II S.-D. Chen and A.R. Ramli,Minimum mean brightness error bi-histogranll equalization in contrast enhanceMent, IEEE Trans double-core Cpu and 4B of ram using matlab 2012b. The Consum Electron., vol. 49, no. 4, pp. 1310-13 19, Nov. 200 simulation results show that UCIQE has the fastest execution [12] C. Wang and Z. Ye, " Brightness preserving histogram equalization with maximum entropy: A variational perspective, IEEE Trans. Consum speed. The CiQI measure requires 4 times running time than Electron,vo.51,no.4,pp.1326-1334,Nov.2005 the UCIQE metric although they all combine colourfulness 113]C. H. Ooi, N.S. P. Kong, and H. Ibrahim, "Bi-histogram equalization nd contrast metrics. This is useful for real -time underwater with a plateau limit for digital image enhancement, IEEE Trans Consum. Electron., vol 55, no. 4, pp. 2072-2080, NoV. 2009 applications [14 B. Bringier, N. Richard, M.-C. Larabi, and C. Fernandez-Maloigne o-reference perceptual quality assessment of colour image,in PI Eur Signal Process. Conf(EUSIPCO), Sep 2006, pp. 1-5 VI CONCLUSION [15 A. Maalouf and M.-C.Larabi, A no reference objective color image A first-of-kind underwater colour image quality evaluation sharpness metric, in Proc. Eur: Signal Process. Conf. (EUSIPCOJ Aug.2010,Pp.1019-102 metric is proposed. The approach extracts the most relevant [16] D. Hasler and S. E. Suesstrunk, "Measuring colorfulness in natural CIELab space statistical features that are representative for images, Proc. SPIE, vol. 5007, pp 87-95,Iun. 2003 underwater image degradations such as colour cast, blurring [17 Y.-Y. Fu,"Color image quality measures and retrieval, Ph. D. dissertation, Dept. Comput. Sci., New Jersey Inst. Technol and noise caused by attenuation, floating particles and Newark. NJ USA. Jan 2006 lighting. The results indicate that the proposed metric has [I8] k Panetta, C. Gao, and s. Agaian, "No reference color image contrast fast processing time, which makes it applicable for real-time pp.643-651,Aug2013 Image processing It is able to successfully predict the relative [19] S. S. Agaian, B. Silver, and K.A. Panetta,"Transform coffi distortion with similar scenes and the difference between cient histogram-based image enhancement algorithms using contrast enhancement results It also shows better correlation with entropy, IEEE Trans. Image Process, vol. 16. no. 3. pp. 741-758 Ma.2007 subjective evaluation. The proposed approach is promising [20)KAPanetta, E.J.Wharton, and S.S.Agaian, "Human visual system in terms of both computational efficiency and practical based image enhancement and logarithmic contrast measure, IEEE reliability for real-time applications and most importantly it is Trans. Syst., Man, Cybern. B. Cybern, vol. 38, no. 1. pp. 174-188 Feb.2008 a meaningful structural model to realize effective underwater [21] sS. Agaian, K Panetta, and A M Grigoryan, "A new measure of image colour image quality evaluation for different applications nhancement, in Proc. IASTED Int. Conf. Signal Process. Commun 2000.pp.19-22. ACKNOWLEdGmEnt [22 P. Marziliano, F. Dufaux, s. Winkler, and T. ebrahimi, A no-reference pcrccptual blur metric, " in Proc. IEEE Int. Conf. Image Process, vol. 3 The authors would like to thank Z. Wei and B. Zheng 2002,Pp.I157-IL60 O Department of Information, Ocean University of China. [23] E. Ong. W Lin, Z. Lu. S. Yao, X. Yang, and L Jiang, " No-reference Qingdao, Shandong, China, for providing the tank images JPEG-2000 image qualily metric, in Proc. IEEE Int. Conf. Multimedia Expo,vol.1.Ju.2003,pp.I5451548 [24] J. Caviedes and F Oberti, "A new sharpness metric based on local kur REFERENCES tosis, edge and energy information,Signal Process, Image Commun vdl.19,n0.2,pp.147-161,Feb.200 [I] M. Yang and z. Ji, "Underwater color image enhancement based on [25] Y. Y Schechner and N Karpel, "Recovery of underwater visibility and quaternion and fuzzy morphological sieves, Chin. J. Sci. Instrum structure by polarization analysis, IEEE J. Ocean. Eng, vol. 30, no. 3, Vol.33,n.7,pp.1601-1605,2012 pp.570-587,Jul.2005 [2] M. Yang and C.L. Gong, "Underwatcr imagc restoration by turbulence [26] W. Hou, A.D. Weidemann, D. J. Gray, and G.R.Fournier model based on image quality evaluation, " Int J Advancements Comput Imagery-derived modulation transfer function and its applications Technol. vol. 4. no. 15, pp 38-46. 2012 for underwater imaging, PrOc. SPIE, vol. 6696, p. 669622 [3 M. Yang and ZJi, Quaternion color morphological filters for under- Aug.2007 watcr images, Adv. Sci. Lett., vol. 7, no. 5, pp. 557-560, 201 [4]R. Schettini and S. Corchs, "Underwater image processing: state of the [27 A. Arnold-Bos, J. P. Malkasse, and G. Kervern, Towards a model-free denoising of underwater optical images, in PrOC. IEEE Eur: Oceans art of restoration and image enhancement methods, FURASP. Adv Conf, vol. I. Brest, France, Jun. 2005, pp. 527-532 Signal Process., vol. 2010, p. 14, Jan. 2010 [28] W.K. Pratt, Digital Image Processing. New York, NY, USA: Wiley [5]R.A. Pramunendar, G. F. Shidik, C. Supriyanto, P. N. Andono, and M. Hariadi, "Auto level color correction for underwater image matching [29] M. Arredondo and K. Lebart,"A methodology for the systematic optimization, IntJ Comput. Sci. Netw. Secur, vol. 13, no. l, pp. 18-23 assessment of underwater video processing algorithms, in Proc. IEEE 2013 Eur: Oceans Conf, voL. 1. Jun. 2005, pp. 362-367 [6] C. Ancuti, C. O. Ancuti, T. Haber, and P. Bekaert, "Enhancing under- [30] M. Yang and A. Sowmya, ""New image quality evaluation metric water images and videos by fusion, in Proc. CVPR, Jun. 2012 for underwater video, " IEEE Signal Process. Lett., vuL. 21, [10. 10 pp.1215-1219,Oct.2014. [7] Z. Wang, A. C. Bovik. H. R. Sheikh, and E. P. Simoncelli,"Image [31] J. Y. Chiang and Y.C. Chen, "Underwater image enhancement by quality assessment: From error visibility to structural similarity,"IEEE wavelength compensation and dehazing IEEE Trans. Image Process Trans. Image Process, vol. 13, no. 4, pp. 600-612, Apr. 2004 l.21,no.4,pp.1756-1769,Apr.201 [8] C Gao, K. Panetta, and S. Agaian, "A new color contrast enhancement [32] D. Akkaynak, E. Chan, J J. Allen, and R. T. Ilanlon, "Using spectrom algorithm for robotic applications, in ProC. IEEE Conf. Technol. Prac etry and photography to study color underwater, in PrOc. OCEANS, tical Robot Appl. (TePRA), Apr. 2012, pp 42-47 Waikoloa, HI, USA. Sep. 201l, pp. 1-8 YANG AND SOWMYA: UCIQE METRIC 071 33]I. Kjerstad,"Underwater Imaging and the effect of inherent optical Arcot Sowimlya received the Ph. D. degree in properties on image quality, M..thesis, Dept. Biol., Norwegian Univ computer science from IIT Bombay, besides other Sci. Technol., Trondheim, Norway, 2014 degrees in mathematics and computer science d [34]R J Davies-Colley, Measuring water clarity with a black disk. " Limno She is currently a Professor with the school of Oceanogr, vol 33, n0. 4. Pp 616-623, 1988 Computer Science and Engineering, University of [35] VQEG.(Aug. 2003 ). Final Report From the video Quality Experts New South Wales, Sydney. Her research has been Group on the validation of Objective Models of video Quality Assess- applied to extraction of linear teatures in remotel ment[onLine].Available:http:/www.vqeg.org sensed iNages and Teature extractioN, recognition [36 K. He, J. Sun, and X. Tang,"Guided image filtering, IEEE and computer aided diagnosis in medical images. Trans. Pattern Anal. Mach. Intell., vol. 35, no. 6, pp. 1397-1409 Her areas of research include learning in vision for Jun.2013 segmentation and object recognition, and embedded [37]R Fattal, "Single image dehazing ACM Trans. Graph, vol 27, no 3, system design 008 [38]J.P. Tarel and N. Hautiere, " Fast visibility restoration from a single color or gray level image, in PrOc. IEEE Conf. Comput. Vis. Pattern Recognit., Sep /Oct. 2009, pp. 2201-2208 [39]K. Iqbal, S.R. Abdul, M. Osman, and A. Z. Talib, "Underwater image enhancement using an integrated colour model, Int J. Comput. Sci. vol.32,no.2,pp.239-244,2007. Y M'12 Haerbin China in 1978. She received the b.s. and m.s. degrees in electronics engineering from LanZhou University, Gansu, China, in 2004, and the ph. d. dcgrcc in nformation science and engineering from the Ocean University of China, Qingdao, in 2009 She was a pusl-Doctoral Fellow with the intermet of Things Engineering Department, Jiangnan University, China, from 2010 to 2013. Since 2009 she has been an associate professor with th Elcctronic Enginccring Dcpartmcnt, Huaihe Institute of Technology. She was a Visiting Scholar with the School of Computer Science and Engineering, University of New South Wales, Sydney, Australia, fruIn 2013 to 2014. She has authored over 30 arlicles and holds two patents. Her research interests include underwater vision, image processing, computer vision, and 3D reconstruction 【实例截图】
【核心代码】