• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Fuzzy image restoration using modulation transfer function①

    2016-12-29 05:34:30YuanYuanYiYaohuaLiuJuhua
    High Technology Letters 2016年4期

    Yuan Yuan (袁 媛), Yi Yaohua, Liu Juhua

    (School of Printing and Packaging, Wuhan University, Wuhan 430079, P.R.China)

    ?

    Fuzzy image restoration using modulation transfer function①

    Yuan Yuan (袁 媛), Yi Yaohua②, Liu Juhua

    (School of Printing and Packaging, Wuhan University, Wuhan 430079, P.R.China)

    The key difficulty of restoring a fuzzy image is to estimate its point spread function (PSF). In the paper, PSF is modelled based on modulation transfer function (MTF). The first step is calculating the image MTF. In the traditional slanted-edge method, a sub-block is always manually extracted from original image and its MTF will be viewed as the result of the whole image. However, handcraft extraction is inefficient and will lead to inaccurate results. Given this, an automatic MTF computation algorithm is proposed, which extracts and screens out all the effective sub-blocks and calculates their average MTF as the final result. Then, a two-dimensional MTF restoration model is constructed by multiplying the horizontal and vertical MTF, and it is combined with conventional image restoration methods to restore fuzzy image. Experimental results indicate the proposed method implementes a fast and accurate MTF computation and the MTF model improves the performance of conventional restoration methods significantly.

    fuzzy image, modulation transfer function (MTF), slanted-edge method, automatic computation, MTF restoration model

    0 Introduction

    Fuzzy images are very common in the digital image processing field and their restoration is of great significance because blurriness in the image not only influences understanding or interpretation for human observers, but also subsequent numerous applications. The essence of restoring fuzzy images is to figure out point spread function (PSF) of the images, which is also the key difficulty to be solved in traditional fuzzy image restoration algorithms. However, PSF is difficult to be estimated in spatial domain since the fuzzy images always suffer from different and complex processing and handling. In the theory of image restoration, PSF and MTF are the expressions of the same physical quantity in spatial and frequency domains respectively, which can be converted to each other by two-dimensional Fourier transform and inverse Fourier transform. Therefore, it is a solution for estimating the restoration model for fuzzy images based on MTF. Then the model can be applied to traditional image restoration algorithms and the fuzzy images can be restored.

    The MTF-based fuzzy image restoration algorithms are widely studied in remote sensing images[1-4]. However, researchers mainly focus on the applications and adjustments of MTF to improve the effects of image restoration, while pay little concern about the calculation of MTF values. In Refs[5-7], MTF was combined with conventional restoration algorithms to increase the sharpness of original images. Exponential methods were put forward to stretch MTF curve of images to increase the contrast and clarity[8]. Ref.[9] raised the ability of noise resisting of MTF extraction via enhancing the local image by homomorphic filter algorithm. To further optimize the quality of restored images, MTF was integrated with other algorithms (such as super resolution) to deal with image deburring, and results showed that the estimated precision of PSF was greatly improved[10]. At present, the computation algorithms of image MTF are mainly divided into three categories according to different shape features of the sub-blocks extracted from the image: pinhole image algorithm, slit-image algorithm and slanted-edge algorithm[11-14]. Compared with the other two algorithms, the slanted-edge algorithm is most widely used in practical applications for the following reasons: 1) slanted edges are very common in digital images, leading to a great possibility to obtain slanted-edge sub-blocks for MTF computation; 2) the energy of a slanted edge is larger than that of a pin-hole or a slit, making the MTF results more accurate and robust. However, the existing slanted-edge algorithm has two disadvantages. Firstly, the slanted-edge sub-blocks are usually extracted from images by manual handling and operations, which is inefficient and time-consuming. Secondly, in conventional slanted-edge method, only one sub-block is manually extracted and its MTF is regarded as the result of whole image. Such handcraft extraction is inefficient and will lead to inaccurate results.

    In this work, the above two problems are solved by proposing an automatic MTF computation algorithm with the following procedures:

    (1) Extracting sub-blocks from original image automatically;

    (2) Adopting and improving the traditional slanted-image method for MTF computation for each sub-block;

    (3) Averaging the MTF values of all the sub-blocks as the final MTF result of the fuzzy image.

    Then, a two-dimensional MTF restoration model will be constructed and combined with the traditional image restoration algorithms to recovery the fuzzy image.

    The rest of this paper is organized as follows. In Section 1 an improved automatic MTF computation algorithm is proposed. Then the MTF-based restoration model is constructed and integrated into the conventional fuzzy image restoration methods in Section 2. Section 3 presents two kinds of comparison experiments on MTF computation and fuzzy image restoration. Final remarks are contained in Section 4.

    1 Automatic MTF computation

    The complete process of fuzzy image restoration using MTF is shown in Fig.1.

    Fig.1 Flowchart of the process of fuzzy image restoration using MTF

    1.1 Restricted conditions of sub-blocks

    Successful MTF computation for slanted-edge methods requires that selected sub-blocks with edges are limited by some specific conditions. Fig.2 shows the acceptable sub-blocks with slanted edges, which include both vertical edges and horizontal edges.

    Fig.2 Acceptable sub-blocks with slanted edges

    (1) Contrast

    If the slanted-edge sub-block is of low contrast, it means that the differences across edge are small, it will always give rise to unreliable MTF results. Therefore, the contrast of sub-block should be limited. According to experimental results[15], the threshold of image contrast is 0.2.

    (2) Edge angle

    Edge angle is defined as the angle between the edge and the positive horizontal axis across the lower end of the edge. If the edge angle is too few or too large, there will be few lines across edges, leading to errors in constructing resampled points for the edge. Experiments are put forward to figure out the appropriate edge angles. Since (0, 90] and [90, 180) are symmetrically distributed, different edge angles in (0, 90] are tested and then the corresponding angles in [90, 180) can be obtained.

    Three MTF-based metrics, namely feature frequency (f_fre), feature MTF (f_MTF) and MTF Area (MTFA), are adopted to evaluate the MTF computation results of sub-blocks with different edge angles, all of which follow the same law that the larger they are, the better the image quality is. Fig.3 shows the computation results of three metrics of different sub-blocks. Observed from the distributions of three curves, it can be deduced that the optimal edge angle range is (0 30] and [60, 90).

    Fig.3 Distributions of three MTF-based metrics of sub-blocks with various edge angles

    (3) Edge size

    Different edge sizes have little influence on the computation accuracy of MTF. However, when the edge size increases, the sub-block becomes larger as well, which will increase computational capacity. Furthermore, if the edge is too long, it might not be able to extract suitable regions as sub-blocks from fuzzy images with various image sizes. According to previous studies[15], the appropriate edge size is about 128 pixels, with the corresponding sub-block size of 80×100.

    1.2 Automatic extraction of sub-blocks

    (1) Edge detection. The premise of sub-block extraction is to find out straight lines in the image. Hough transform is an efficient method for line detection since it is of high speed and precision, and easy to implement. The detected line segments are the potential edges of sub-blocks.

    (2) Edge screening. According to the three limited conditions of sub-blocks discussed in Section 1.1, the line segments can be screened out.

    (3) Extraction of sub-blocks. A sub-block is a rectangle with slanted edges in it. With proper edges obtained, the region of sub-blocks can be extracted by obeying one basic rule: the short sides of the rectangle should both intersect the edges. To simplify the extraction of the rectangle, two ends of the edge are taken as the center point of the short sides. Then both vertical sub-blocks and horizontal sub-blocks can be extracted.

    1.3 MTF computation for each sub-block

    The slanted-edge algorithm is employed to calculate MTF for each sub-block. The detailed steps are described as follows.

    (1) Denoising preprocessing. When restoring the image in frequency domain, noises (high-frequency signals) will be amplified. Hence, self-adapting median filtering is conducted to reduce the noises in the fuzzy image.

    (2) Sub pixel location and edge fitting. For each line across the edge, the location of pixel is found which has the biggest gray scale change and get the positions of its neighboring three pixels. Then, a cubic polynomial model can be established (in Eq.1) and the location of sub pixel can be determined since the second derivative of sub pixel equals 0 (in Eqs(2) and (3)). Then the sub pixel line can be fitted with a polynomial curve fitting method.

    f(x)=a(1)x3+a(2)x2+a(3)x+a(4)

    (1)

    f″(x)=6a(1)x3+2a(2)

    (2)

    x=-a(2)/(3×a(1))

    (3)

    where x refers to the locations of four pixels (columns for vertical edges and lines for horizontal edges) and f(x) is the gray scale. a(1), a(2), a(3) and a(4) are the coefficients to be solved with known x and f(x).

    (3) Pixel resampling and edge spread function (ESF) fitting. According to the fitted sub pixel line, all pixels in fuzzy image are projected and resampled to obtain more points for accurate MTF computation, so it is with their corresponding gray scales. Then the Fermi function (in Eq.(4)) is used to fit the ESF curve because of its superior fitting effect and strong ability to suppress noises.

    f(x)=a/(1+e)

    (4)

    where x is the locations of resampled points and f(x) is their corresponding gray scales, a expresses the amplitude, c reflects the steepness of the ESF curve, and b is the horizontal axis of the symmetric center of the curve. They can be figured out with known x andf(x).

    (4) Line spread function (LSF) computation. LSF is the first order difference of ESF (in Eq.(5)). However, experimental results shows that the number of effective MTF values is not enough and might lead to errors. To increase the number of MTF values, the zero padding method is adopted to the both ends of the LSF curve, as is shown in Fig.4.

    LSF(n)=ESF(n)-ESF(n-1)

    (5)

    where n is the number of resampling points.

    (5) MTF computation. Perform Fourier transform to LSF curve and normalize all the results by the largest magnitude of the first point.

    MTF(n)=|DFT(LSF(n))|

    (6)

    norm_MTF(n)=MTF(n)/MTF(1)

    (7)

    where norm_MTF(n) means the normalized MTF values.

    Fig.4 contrast of MTF before and after zero padding to LSF curves

    1.4 Computation of MTF for overall image

    Since the extracted sub-blocks consist of vertical and horizontal edges, the final MTF result is made up of two components: vertical MTF vector and horizontal MTF vector. They are calculated by firstly performing the slanted-edge method on all the sub-blocks of two directions and then averaging the whole MTF values to get the final MTF result.

    2 MTF-based fuzzy image restoration

    The construction of restoration model is the main problem to be solved in conventional image restoration methods. In this paper, a two-dimensional restoration model based on MTF values is established by multiplying the vertical MTF vector and the horizontal MTF vector of the fuzzy image (in Eq.8).

    MTF(u,v)=MTFu×MTFv

    (8)

    where u and v refer to vertical and horizontal directions, MTFuand MTFvare vertical and horizontal MTF vectors.

    When constructing a two-dimensional MTF model, the symmetrical mapping method is adopted, namely, MTF values in the MTF model are the same in the range of equal-length radius with the model center as the center of a circle. The MTF values of frequencies in [0.5, 1] are close to 0 and thereby have little impact on constructing the MTF model used to describe the point spread in fuzzy images. As a result, only the MTF values in [0, 0.5] are employed.

    The size of MTFu,v, which is determined by the sizes of vertical and horizontal MTF vectors, is much smaller than that of the fuzzy image to be restored. Therefore, MTFu,vis interpolated with the bicubic interpolation algorithm so that it can be applied to traditional image restoration algorithms. In this paper, three image restoration methods are adopted to recover the fuzzy image: Wiener filtering (referred to as ‘WN’) algorithm, constrained least square (referred to as ‘CLS’) filtering and regularized Lucy-Richardson (referred to as ‘RLR’) algorithm.

    3 Experiments

    3.1 Evaluation metrics

    Objective evaluation metrics are adopted to measure the performance of image restoration methods. Three types of metrics, which contain seven indicators, are employed in the paper.

    (1) MTF based metrics

    The f_fre, f_MTF and MTFA are used to evaluate the clarity for restored images. Larger indicators suggest a better image quality.

    (2) Gray scale metrics

    ① Signal noise ratio (SNR). It measures the amounts of noises in the images[16]. Larger SNR indicates fewer noises the image has.

    (9)

    ② Gray mean grads (GMG). It describes the definition of image and sensitively reflects the contrast in details and texture properties[16]. Larger GMG means higher image clarity.

    GMG=

    (10)

    ③ Contrast. It has the same effect as the definition with respect to image quality. The larger the contrast is, the deeper the textures are, and the clearer the visual effects are[16].

    (11)

    (3) HVS metric

    Just noticeable difference quality loss (JNDQL) describes the degeneration degrees of image quality perceived by human visual systems. The smaller it is, the better the image quality is. The implementation process is as follows:

    First, calculating the perceptual acutance of the image using contrast sensitivity function (CSF):

    (12)

    (13)

    where v refers to the spatial frequency, a=75, b=0.2, c=0.8 and K=34.05[17].

    Second, researches indicate there exists an objective metric (OM)[17]which has a certain relation with perceptual acutance. When acutance is less than 0.8851, the smaller OM is, the clearer the image will be. While it is greater than 0.8851, the perceptual image will be very approximate to pristine image, and the visual quality is no longer linear to acutance.

    (14)

    Third, calculating the JNDQL based on OM:

    (15)

    3.2 MTF computation experiment

    To verify the effectiveness and accuracy of the proposed automatic MTF computation method, a publicly available computation software, which is sfrmat[18], is adopted to conduct the contrast experiment. Sfrmat software is a Matlab tool that analyzes the spatial frequency response (SFR) from an image which contains a slanted edge. It is developed to measure the resolution for electronic cameras according to standard ISO 12233.

    In this experiment, both the proposed method and sfrmat are used to calculate the MTF values of the 515×600 pixel standard test image for resolution measurements. Both methods are performed with Matlab R2014a on computers with Intel Core i7-4770 3.4GHz and 8GB RAM.

    As is shown in Fig.5, the proposed method automatically extracts 14 sub-blocks from the standard test image, including five vertical and nine horizontal sub-blocks. In Fig.6(a) and (b), the MTF calculation results of all the sub-blocks in two directions are displayed and the two black curves in bold are their averages. From the distributions of MTF values, it is obvious to find that there exist some abnormal values and the computation errors caused by them can be minimized by calculating the averages of all the MTF values. With respect to sfrmat, only two sub-blocks, one vertical and one horizontal sub-block, are manually extracted and used to calculate the SFR values with sfrmat software.

    Fig.5 Standard test image for resolution measurements and extracted sub-blocks by proposed method

    Fig.6(c) shows distributions of SFR results by sfrmat and the average MTF results by the proposed method. The differences between them are minor. Besides, feature MTF and feature frequency of the two curves are very close. The main causes for the small difference are the different number of resample points and different fitting functions adopted in the two algorithms.

    (a) Vertical MTF values

    (b) Horizontal MTF values

    (c) Comparison between sfrmat and the proposed method

    Table 1 shows the average MTF computation time of one sub-block by sfrmat and the proposed method. The time of sfrmat is figured out by calculating the statistical average of 20 tests. The proposed automatic MTF computation method costs much less time than sfrmat. This is because the proposed method is fully automatic computation while sfrmat requires that we manually select appropriate sub-blocks.

    As is shown in this comparison experiment, despite the minor calculation differences, the calculation accuracy of the proposed automatic MTF computation method is comparable to sfrmat. In addition, the proposed method has the advantages of fast and automatic MTF computation, demonstrating its effectiveness and reliability in practice.

    Table 1 Time consumed by calculating the MTF values of a sub-block in the standard test image using two methods

    MethodsAveragetimepersub?block(s)Sfrmat15.84ProposedautomaticMTFcomputationmethod4.21

    3.3 Fuzzy image restoration experiment

    To evaluate the performance of the two-dimensional MTF restoration model in image restoration algorithms, three groups of comparison experiments are carried out on three fuzzy images, including a natural scene (NS) image (in Fig.7(a)) from LIVE database[19], a remote sensing (RS) image (in Fig.8(a)) and an image made by adding Gaussian noises to the RS image (GNRS) (in Fig.9(a)). They are restored by the following restoration methods, the conventional image restoration methods (‘WN’, ‘CLS’, ‘RLR’), their corresponding MTF-based methods (referred to as ‘M-WN’, ‘M-CLS’ and ‘M-RLR’), and a blind deconvolution (referred to as ‘BD’) algorithm[20-22].

    Fig.7 Restoration results of the NS image

    The NS and RS images are tested to validate effectiveness of MTF model to conventional restoration methods. The GNRS image is used to test the anti-noise ability of the restoration methods. For clear observations on image details, the same local regions of GNRS image and its restored images are shown in Fig.9(b)~(i). Besides, the evaluation results are listed in Table 2, Table 3 and Table 4, respectively. The best three values of each indicator are shown in bold.

    Fig.8 Restoration results of the RS image

    Fig.9 Restoration results of local regions of the GNRS image

    Fig.7 and Fig.8 indicate that, from the perspective of human visual perception, the restoration effects of algorithms using MTF models are superior to the conventional ones. Take Fig.7 for example, the edges of the texts on the plane in Fig.7(f), (g) and (h) become very sharp and the details of grasses on the ground are clearly represented after restoration with MTF-based methods. However, the results of the other conventional algorithms show small improvements in visual effects. The textures in Fig.7(b), (c), (d) and (e) are relatively unclear and their contrast seems insufficient. The subjective judgment can be confirmed by the quantitative image quality evaluation metrics in Table 2 and Table 3. All the metrics of M-WN and M-CLS achieved the best three values, which demonstrates significant effectiveness of MTF models in promoting the quality of fuzzy images for WN and CLS. However, it has little influence on RLR algorithm.

    Fig.9 shows the restoration results of local regions of GNRS image. For the NS image and RS image which have no noises, the restoration results of M-WN and M-CLS are very close from both subjective and objective evaluations. But, for the GNRS image, it seems that CLS and M-CLS achieved better performance than M-WN according to the evaluation indicators in Table 4. However, as shown in Fig.9, it is not consistent with the subjective observations that M-WN is superior to both CLS and M-CLS. The reason for this phenomenon is that, compared to M-WN algorithm, the Gaussian noises in the GNRS image are greatly increased in the recovering process of CLS and M-CLS. As a consequence, M-WN has a stronger anti-noise performance than CLS and M-CLS. Besides, RLR shows a good ability of restraining noise reflected by higher SNR values as well.

    All in all, the MTF restoration model greatly improves the restoration performance by conventional restoration methods. Among them, the MTF-based wiener filtering algorithm has the best performance in both restoration effects and noise suppression.

    Table 2 Quality evaluation results of the natural scene image

    *Best three values of each indicators are shown in bold.

    Table 3 Quality evaluation results of the RS image

    *Best three values of each indicators are shown in bold.

    Table 4 Quality evaluation results of the GNRS image

    *Best three values of each indicators are shown in bold.

    4 Conclusions

    This study presents an improved fuzzy image restoration approach based on the image MTF. This approach firstly improves the conventional slanted-edge method by automatically extracting multiple sub-blocks from original image and calculating the average MTF values to reduce the abnormal errors, and secondly recoveries the fuzzy image by establishing the two-dimensional MTF restoration model and combining it with the conventional image restoration algorithms. According to the comparison experimental results of MTF computation and fuzzy image restoration, the improved automatic slanted-edge algorithm achieves a rapid and accurate MTF calculation for fuzzy images, and the MTF model improves the performance of conventional restoration algorithms significantly. Since not all the digital images are equipped with slanted edges, researching on sub-blocks with other shape features can provide more solutions for automatic MTF computation in the future work.

    [ 1] Wang H N, Zhong W, W Jing, et al. Research of measurement for digital image definition. Journal of Image and Graphics, 2004, 9(7): 828-831 (In Chinese)

    [ 2] Mu X. Remote Sensing Image Restoration Based On Spatial Optics Remote Sensor MTF Compensation: [Ph.D dissertation]. Changchun: Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, 2012. 16 (In Chinese)

    [ 3] Li T C. Research On Modulation Transfer Function Calculation and Remote Sensing Image Restoration Based On Slanted-edge Method: [Ph.D dissertation]. Hangzhou: College of Optical Science and Engineering, Zhejiang University, 2011. 20-21 (In Chinese)

    [ 4] Zeng Y, Chen S P, Yu J. Modulation transfer function compensation technology of optical remote sensing imaging system. Chinese Space Science and Technology, 2010, 30(4): 38-43 (In Chinese)

    [ 5] Gu X F, Li X Y, Min X J, et al. CBERS-02 satellite CCD camera MTF on-orbit measurement and MTF compensation. Science in China (Series E: Technological Sciences), 2005, 35: 26-40 (In Chinese)

    [ 6] Reimann D A, Jacobs A, Holly A. Use of wiener filtering in the measurement of the two-dimensional modulation transfer function. In: Proceedings of SPEI 3977: Physics of Medical Imaging, San Diego, USA, 2000. 670-680

    [ 7] Wang C, Gu X F, Yu T, et al. On-orbit MTF measurement and image restoration for ZY-3 satellite. Computer Engineering, 2014, 40(4): 237-241 (In Chinese)

    [ 8] Wang X H, Qiao Y L, Hong J. A technique of RS image restoration and in-flight measurement of remote sensor MTF. Remote Sensing Technology and Application, 2006, 21(5): 440 (In Chinese)

    [ 9] Meng W, Jin L X, Li G N, et al. Application of MTF in remote sensing image restoration. Infrared and Laser Engineering, 2014, 43(5): 1690-1696 (In Chinese)

    [10] Liu Y Y, Jin W Q, Su B H. Identification of blurred system based on super-resolution reconstructing image schemes. Journal of Optoelectronics Laser, 2005, 16(2): 213-215 (In Chinese)

    [11] McGillem C D, Anuta P E, Malaret E, et al. Estimation of a remote sensing system point spread function from measured imagery. In: Proceedings of the Machine Processing of Remotely Sensed Data, West Lafayette, USA, 1983. 81

    [12] Dominique L, Joel D, Francoise R. MTF measurement using spotlight. In: Proceedings of the International Geoscience and Remote Sensing Symposium (IGARSS), Pasadena, USA, 1994. 2010-2012

    [13] Cao X H, Huang H K, Linaglou S. A novel algorithm for measuring the MTF of a digital radiographic system with a CCD array detector. International Society for Optics and Pho-tonics, 2000, 3922: 580-589

    [14] Dennis D L, Helder T, Choi. IKONOS Satellite in Orbit Modulation Transfer Function (MTF) Measurement Using Edge and Pulse Method: [Ph.D dissertation]. Brookings: Electrical Engineering Department, South Dakota State University, 2002. 45

    [15] Yuan Y, Yi Y H, Miao M J. An automatic calculation method of MTF and the application in blurred images restoration. Applied Mechanics and Materials, 2015, 731: 141-146

    [16] Jia Y H. In: Digital Image Processing. Wuhan: Wuhan University Press, 2006. 182-184

    [17] International Imaging Industry Association. P1858 Standard for Camera Phone Image Quality. New York, USA: IEEE Standards Association, 2012

    [18] Estribeau M, Magnan P. Fast MTF measurement of CMOS imagers using ISO 12233 slanted-edge methodology. In: Proceedings of SPIE - The International Society for Optical Engineering, Saint-Etienne, France, 2003. 243-252

    [19] Sheikh H R, Sabir M F, Bovik A C. A statistical evaluation of recent full reference image quality assessment algorithms. IEEE Transactions on Image Processing, 2006, 15: 3440-3451

    [20] Gonzalez R C. Digital Image Processing Using MATLAB. Beijing: Publishing House of Electronics Industry, 2005.126-133

    [21] Yan H, Yan W J, Li W W. Image restoration based on Lucy-Richardson algorithm. Computer Engineering, 2010, 36(15): 204-205 (In Chinese)

    [22] Jia H P. Blind deconvolution algorithm in application of image restoration. Information Technology, 2011, 5: 38-39 (In Chinese)

    Yuan Yuan, born in 1989. She has been working toward her Ph.D degree with Professor Yi Yaohua at the School of Printing and Packaging of Wuhan University. She received her M.S. degree in computer science and technology from Wuhan University, Wuhan, Hubei province, China, in 2011. Her current research interests include image quality assessment and image processing.

    10.3772/j.issn.1006-6748.2016.04.016

    ① Supported by the National High Technology Research and Development Programme of China (No. 2012AA12A305), the National Key Technology R&D Program of the Ministry of Science and Technology (No. 2013BAH03B01), Fundamental Research Funds for the Central Universities of China (No. 2042015kf0059) and China Postdoctoral Science Foundation (No. 2015M582277).

    ② To whom correspondence should be addressed. E-mail: yyh@whu.edu.cn Received on Nov. 10, 2015,

    精品人妻视频免费看| 国产私拍福利视频在线观看| 99热这里只有是精品在线观看| 亚洲精品一卡2卡三卡4卡5卡| 变态另类丝袜制服| 日本在线视频免费播放| 日韩一本色道免费dvd| 插阴视频在线观看视频| 欧美潮喷喷水| 久久亚洲精品不卡| 毛片女人毛片| 国产在线男女| 国内精品久久久久精免费| 真人做人爱边吃奶动态| 亚洲国产精品久久男人天堂| 99热这里只有精品一区| 国产蜜桃级精品一区二区三区| 国产视频内射| 老熟妇仑乱视频hdxx| 国模一区二区三区四区视频| 熟妇人妻久久中文字幕3abv| 国产黄a三级三级三级人| 99久久成人亚洲精品观看| 美女免费视频网站| 简卡轻食公司| 久久久欧美国产精品| 一级a爱片免费观看的视频| 久久久精品94久久精品| 久久久久九九精品影院| 在线观看一区二区三区| 久久精品国产亚洲av天美| 少妇人妻一区二区三区视频| 可以在线观看的亚洲视频| 中国美女看黄片| 在线天堂最新版资源| 少妇猛男粗大的猛烈进出视频 | 91精品国产九色| 日日啪夜夜撸| 最近中文字幕高清免费大全6| 国产精品电影一区二区三区| 精品乱码久久久久久99久播| 高清毛片免费看| 一a级毛片在线观看| 久久精品国产99精品国产亚洲性色| 国产三级中文精品| 麻豆国产av国片精品| 你懂的网址亚洲精品在线观看 | 在线免费观看的www视频| 不卡一级毛片| 麻豆一二三区av精品| 国产精品一二三区在线看| 欧美在线一区亚洲| 熟女电影av网| 99久久精品国产国产毛片| 亚洲欧美精品自产自拍| 男人和女人高潮做爰伦理| av.在线天堂| 久久午夜亚洲精品久久| 免费观看在线日韩| 给我免费播放毛片高清在线观看| 久久久久国产精品人妻aⅴ院| av女优亚洲男人天堂| 国产精品美女特级片免费视频播放器| 男人的好看免费观看在线视频| 免费不卡的大黄色大毛片视频在线观看 | 国产高清视频在线播放一区| 国产视频内射| 欧美bdsm另类| 午夜久久久久精精品| 亚洲av美国av| 国产伦精品一区二区三区四那| 变态另类丝袜制服| 国产亚洲精品综合一区在线观看| 我要看日韩黄色一级片| 亚洲在线自拍视频| 精品久久久久久久久av| 免费av观看视频| 日本免费a在线| 99久久九九国产精品国产免费| 精品久久久久久久末码| av.在线天堂| 最近的中文字幕免费完整| 亚洲精品日韩在线中文字幕 | 俄罗斯特黄特色一大片| 精品少妇黑人巨大在线播放 | 欧美成人a在线观看| 精品不卡国产一区二区三区| 最近手机中文字幕大全| 国语自产精品视频在线第100页| 色综合亚洲欧美另类图片| 国产精品一区二区免费欧美| 有码 亚洲区| 欧美日韩国产亚洲二区| 国产成人a∨麻豆精品| 成年女人看的毛片在线观看| 可以在线观看的亚洲视频| 国产一区二区三区av在线 | 性色avwww在线观看| 欧美一区二区精品小视频在线| 色视频www国产| 菩萨蛮人人尽说江南好唐韦庄 | 精品久久久久久久人妻蜜臀av| 最近视频中文字幕2019在线8| 内射极品少妇av片p| 18禁裸乳无遮挡免费网站照片| 大型黄色视频在线免费观看| 午夜免费男女啪啪视频观看 | 欧美激情国产日韩精品一区| av卡一久久| av在线蜜桃| 国产精品一二三区在线看| 男女做爰动态图高潮gif福利片| 国产探花在线观看一区二区| 免费看美女性在线毛片视频| 三级男女做爰猛烈吃奶摸视频| 久久久色成人| 97人妻精品一区二区三区麻豆| 国产又黄又爽又无遮挡在线| 亚洲国产欧美人成| 久久鲁丝午夜福利片| 亚洲最大成人手机在线| 国产免费一级a男人的天堂| 麻豆av噜噜一区二区三区| 精品一区二区免费观看| 国产激情偷乱视频一区二区| 精品免费久久久久久久清纯| 亚洲人成网站在线播放欧美日韩| 国产中年淑女户外野战色| 少妇人妻一区二区三区视频| 精品久久久久久久久久免费视频| 超碰av人人做人人爽久久| 熟女电影av网| 一级av片app| АⅤ资源中文在线天堂| 国产毛片a区久久久久| 国产真实乱freesex| 亚洲欧美成人综合另类久久久 | 亚洲中文字幕日韩| 久久久国产成人精品二区| 久久精品国产清高在天天线| 亚洲性久久影院| 十八禁国产超污无遮挡网站| 日日摸夜夜添夜夜添av毛片| 联通29元200g的流量卡| 九九爱精品视频在线观看| a级毛色黄片| 亚洲成人久久爱视频| 中文字幕精品亚洲无线码一区| 国产精品一区二区三区四区久久| 黄色一级大片看看| 美女被艹到高潮喷水动态| 久久精品国产亚洲av天美| 综合色丁香网| 国产精品久久久久久亚洲av鲁大| 99久久成人亚洲精品观看| 亚洲七黄色美女视频| 国内少妇人妻偷人精品xxx网站| 一进一出抽搐动态| 久久欧美精品欧美久久欧美| 亚洲天堂国产精品一区在线| 99久久成人亚洲精品观看| 精品日产1卡2卡| 久久久色成人| 俺也久久电影网| 亚洲婷婷狠狠爱综合网| 一夜夜www| 美女黄网站色视频| 观看美女的网站| 午夜福利18| 国产午夜精品论理片| aaaaa片日本免费| 九九热线精品视视频播放| 悠悠久久av| 国产色婷婷99| 国产亚洲精品久久久com| 国产私拍福利视频在线观看| 一个人免费在线观看电影| 91久久精品电影网| 国产在线精品亚洲第一网站| 久久天躁狠狠躁夜夜2o2o| 久久精品国产自在天天线| 国产午夜精品论理片| 久久久久性生活片| 精品久久国产蜜桃| 国产爱豆传媒在线观看| 欧美一级a爱片免费观看看| videossex国产| 色噜噜av男人的天堂激情| 国产成人freesex在线 | 亚洲最大成人中文| 国产精品伦人一区二区| 国产精品精品国产色婷婷| 男女啪啪激烈高潮av片| 亚洲丝袜综合中文字幕| 最近在线观看免费完整版| 亚州av有码| 精品一区二区三区视频在线观看免费| 久久久久久久久久久丰满| 天天躁日日操中文字幕| 国产伦精品一区二区三区视频9| 国产成人freesex在线 | 亚洲中文字幕日韩| av在线老鸭窝| 毛片女人毛片| 亚洲18禁久久av| 国产一级毛片七仙女欲春2| av在线老鸭窝| 午夜精品在线福利| 国产亚洲91精品色在线| 国产精品综合久久久久久久免费| 大又大粗又爽又黄少妇毛片口| 欧美色视频一区免费| 99在线人妻在线中文字幕| 久久精品91蜜桃| 亚洲在线观看片| 成人性生交大片免费视频hd| 欧美一区二区国产精品久久精品| 精品人妻偷拍中文字幕| 国产激情偷乱视频一区二区| 久久久久性生活片| 青春草视频在线免费观看| 免费观看在线日韩| av专区在线播放| 可以在线观看的亚洲视频| or卡值多少钱| 午夜影院日韩av| 黄色配什么色好看| 国内久久婷婷六月综合欲色啪| 久久久久九九精品影院| 天堂√8在线中文| 男女啪啪激烈高潮av片| 国产伦一二天堂av在线观看| 女的被弄到高潮叫床怎么办| 性插视频无遮挡在线免费观看| 国产精品,欧美在线| 国产高潮美女av| 国产伦一二天堂av在线观看| 成熟少妇高潮喷水视频| 欧美成人一区二区免费高清观看| 日韩欧美三级三区| 男女啪啪激烈高潮av片| 精品少妇黑人巨大在线播放 | 久久久久久久久久成人| 欧美高清性xxxxhd video| 毛片女人毛片| 黄色视频,在线免费观看| 看非洲黑人一级黄片| 免费观看在线日韩| 亚洲精华国产精华液的使用体验 | 日韩成人av中文字幕在线观看 | 别揉我奶头~嗯~啊~动态视频| 我要看日韩黄色一级片| 亚洲成人中文字幕在线播放| 婷婷亚洲欧美| 真实男女啪啪啪动态图| 国产麻豆成人av免费视频| 超碰av人人做人人爽久久| 日韩欧美一区二区三区在线观看| 国产精品综合久久久久久久免费| 精品久久久久久久久亚洲| 精品久久久久久久久亚洲| 菩萨蛮人人尽说江南好唐韦庄 | 最近中文字幕高清免费大全6| 日韩欧美三级三区| 精品人妻视频免费看| 亚洲精品日韩av片在线观看| 国产黄色小视频在线观看| 变态另类丝袜制服| 一a级毛片在线观看| 国产女主播在线喷水免费视频网站 | 日本与韩国留学比较| 人人妻人人澡人人爽人人夜夜 | 一区二区三区免费毛片| 精品久久久久久成人av| 国产在视频线在精品| 国产色婷婷99| 成人永久免费在线观看视频| 白带黄色成豆腐渣| av在线蜜桃| 亚洲美女黄片视频| 国产爱豆传媒在线观看| 在线天堂最新版资源| 人妻制服诱惑在线中文字幕| 欧美最黄视频在线播放免费| 国产久久久一区二区三区| 麻豆精品久久久久久蜜桃| 国产亚洲精品久久久com| av视频在线观看入口| a级毛片免费高清观看在线播放| 婷婷六月久久综合丁香| 亚洲精品一区av在线观看| 在线观看免费视频日本深夜| 久久久久久久久久成人| 欧美潮喷喷水| 国产在视频线在精品| 老司机午夜福利在线观看视频| 午夜精品在线福利| 一个人观看的视频www高清免费观看| 麻豆国产97在线/欧美| 久久久久久伊人网av| 美女大奶头视频| 91午夜精品亚洲一区二区三区| 欧美+亚洲+日韩+国产| АⅤ资源中文在线天堂| 简卡轻食公司| 欧美日本视频| 22中文网久久字幕| 俄罗斯特黄特色一大片| 观看美女的网站| 九九热线精品视视频播放| 亚洲自偷自拍三级| avwww免费| 亚洲自拍偷在线| 嫩草影视91久久| 国产私拍福利视频在线观看| 亚洲国产精品成人综合色| 美女xxoo啪啪120秒动态图| 亚洲18禁久久av| 91麻豆精品激情在线观看国产| 3wmmmm亚洲av在线观看| 亚洲乱码一区二区免费版| 日韩av不卡免费在线播放| 亚洲一区二区三区色噜噜| 国产高清不卡午夜福利| 色综合色国产| 亚洲最大成人中文| 欧美日韩国产亚洲二区| 午夜免费男女啪啪视频观看 | 亚洲真实伦在线观看| 男女啪啪激烈高潮av片| 免费在线观看影片大全网站| 亚洲第一区二区三区不卡| 三级男女做爰猛烈吃奶摸视频| 国产精品乱码一区二三区的特点| 国产欧美日韩精品一区二区| 精品久久久久久久久久免费视频| 乱码一卡2卡4卡精品| 亚洲美女搞黄在线观看 | 全区人妻精品视频| 欧美日韩综合久久久久久| 人人妻人人澡人人爽人人夜夜 | 国产三级在线视频| 12—13女人毛片做爰片一| 欧美性猛交黑人性爽| 日韩欧美免费精品| 久久综合国产亚洲精品| 美女xxoo啪啪120秒动态图| 最近视频中文字幕2019在线8| 亚洲无线在线观看| 欧美性猛交黑人性爽| 久久久久久国产a免费观看| 精品久久久久久久久av| 女生性感内裤真人,穿戴方法视频| 亚洲内射少妇av| 女生性感内裤真人,穿戴方法视频| 国产欧美日韩精品亚洲av| 变态另类丝袜制服| 成人鲁丝片一二三区免费| 精品久久久久久久久久久久久| 国产精品永久免费网站| 天堂影院成人在线观看| 夜夜夜夜夜久久久久| 免费看av在线观看网站| 国产v大片淫在线免费观看| 色av中文字幕| 在线播放国产精品三级| 国产一区二区亚洲精品在线观看| 色哟哟·www| 激情 狠狠 欧美| 女人被狂操c到高潮| 国产乱人偷精品视频| 最近最新中文字幕大全电影3| 国产精品一区二区免费欧美| 日韩大尺度精品在线看网址| 国产伦精品一区二区三区四那| 久久久欧美国产精品| 亚洲,欧美,日韩| 久久鲁丝午夜福利片| 在线观看66精品国产| 高清午夜精品一区二区三区 | 日本-黄色视频高清免费观看| 久久久午夜欧美精品| 美女免费视频网站| 国产蜜桃级精品一区二区三区| 99热只有精品国产| 女生性感内裤真人,穿戴方法视频| or卡值多少钱| av视频在线观看入口| 99riav亚洲国产免费| 亚洲欧美日韩高清专用| 午夜激情福利司机影院| 欧美xxxx黑人xx丫x性爽| 国产成人a区在线观看| 午夜福利18| 国产黄色小视频在线观看| 99热这里只有是精品在线观看| 午夜日韩欧美国产| 嫩草影视91久久| 国产精品日韩av在线免费观看| 久久久色成人| 91狼人影院| 国产午夜福利久久久久久| 欧美一区二区国产精品久久精品| 日韩av在线大香蕉| 亚洲国产色片| 久久草成人影院| 最近最新中文字幕大全电影3| 日韩欧美在线乱码| av福利片在线观看| 国产精品综合久久久久久久免费| 欧美日本亚洲视频在线播放| 国产成人91sexporn| 91久久精品国产一区二区三区| 免费人成视频x8x8入口观看| 露出奶头的视频| 高清午夜精品一区二区三区 | 美女 人体艺术 gogo| 免费看日本二区| 欧美色欧美亚洲另类二区| 老司机福利观看| 精品一区二区三区av网在线观看| 中文资源天堂在线| 伦理电影大哥的女人| 欧美日韩综合久久久久久| 日韩av不卡免费在线播放| 久久亚洲精品不卡| 男女边吃奶边做爰视频| 丝袜美腿在线中文| 乱系列少妇在线播放| 晚上一个人看的免费电影| 午夜精品一区二区三区免费看| 搞女人的毛片| 国产女主播在线喷水免费视频网站 | 噜噜噜噜噜久久久久久91| 嫩草影院入口| 99热6这里只有精品| 一本精品99久久精品77| 国产午夜精品论理片| 免费观看在线日韩| 色av中文字幕| 国产单亲对白刺激| 天堂网av新在线| 老熟妇乱子伦视频在线观看| 欧美一区二区精品小视频在线| 成人国产麻豆网| 久久久精品94久久精品| 一级毛片我不卡| 成年女人毛片免费观看观看9| 青春草视频在线免费观看| 成年版毛片免费区| 久久久午夜欧美精品| 日韩欧美三级三区| 毛片一级片免费看久久久久| 人人妻人人看人人澡| 日本黄色视频三级网站网址| 国产真实伦视频高清在线观看| 久久精品影院6| 久久午夜福利片| 男人狂女人下面高潮的视频| 日日摸夜夜添夜夜添小说| 国产老妇女一区| 国产 一区精品| 露出奶头的视频| 1024手机看黄色片| 日韩欧美国产在线观看| 美女免费视频网站| 综合色av麻豆| 最近手机中文字幕大全| 校园春色视频在线观看| 久久久久精品国产欧美久久久| 欧美最黄视频在线播放免费| 国产视频一区二区在线看| 3wmmmm亚洲av在线观看| 俄罗斯特黄特色一大片| 热99re8久久精品国产| 国产成人a区在线观看| 在线a可以看的网站| 最新中文字幕久久久久| 一夜夜www| 亚洲一区高清亚洲精品| 免费av观看视频| 国产乱人视频| 精品久久国产蜜桃| 在现免费观看毛片| 丝袜喷水一区| 九九在线视频观看精品| 日韩欧美一区二区三区在线观看| 日日摸夜夜添夜夜爱| 国产v大片淫在线免费观看| 国产精品99久久久久久久久| 久久精品综合一区二区三区| 中文字幕久久专区| 免费在线观看影片大全网站| 亚洲av美国av| 一个人观看的视频www高清免费观看| 免费看av在线观看网站| 尾随美女入室| 亚洲自拍偷在线| 一级毛片久久久久久久久女| 国产精品一二三区在线看| 国产亚洲精品av在线| 亚洲精品在线观看二区| 亚洲精品国产成人久久av| 午夜激情福利司机影院| 久久久欧美国产精品| 午夜福利在线在线| 久久久精品欧美日韩精品| 亚洲成a人片在线一区二区| 91在线观看av| 日产精品乱码卡一卡2卡三| 国产精品精品国产色婷婷| 成年免费大片在线观看| 精品熟女少妇av免费看| 激情 狠狠 欧美| 中出人妻视频一区二区| 97碰自拍视频| 99在线视频只有这里精品首页| 国产高清不卡午夜福利| 激情 狠狠 欧美| 国产高清三级在线| 亚洲成人中文字幕在线播放| 最近在线观看免费完整版| 国产伦一二天堂av在线观看| 美女黄网站色视频| 久久99热6这里只有精品| 欧美三级亚洲精品| 久久精品影院6| 国产亚洲欧美98| 精品福利观看| 亚洲av免费在线观看| 亚洲成人久久爱视频| 成人永久免费在线观看视频| 99在线人妻在线中文字幕| 淫妇啪啪啪对白视频| 91午夜精品亚洲一区二区三区| 美女xxoo啪啪120秒动态图| 精品99又大又爽又粗少妇毛片| 99视频精品全部免费 在线| 五月玫瑰六月丁香| 国产中年淑女户外野战色| 两性午夜刺激爽爽歪歪视频在线观看| 五月伊人婷婷丁香| 午夜免费激情av| 小蜜桃在线观看免费完整版高清| 日本与韩国留学比较| 亚洲高清免费不卡视频| 可以在线观看毛片的网站| 亚洲美女黄片视频| 国产人妻一区二区三区在| 九九在线视频观看精品| 99国产精品一区二区蜜桃av| 中文资源天堂在线| 亚洲精品粉嫩美女一区| 国产综合懂色| 男人舔女人下体高潮全视频| 欧美色欧美亚洲另类二区| 免费在线观看影片大全网站| 午夜福利在线观看吧| 精品少妇黑人巨大在线播放 | 久久久久性生活片| 成年女人看的毛片在线观看| 国产片特级美女逼逼视频| 一进一出抽搐gif免费好疼| 校园人妻丝袜中文字幕| 亚洲av成人精品一区久久| 色在线成人网| 给我免费播放毛片高清在线观看| 尾随美女入室| 99热只有精品国产| 啦啦啦啦在线视频资源| av.在线天堂| 校园人妻丝袜中文字幕| 黄色欧美视频在线观看| 久久国产乱子免费精品| 国产欧美日韩一区二区精品| 日本熟妇午夜| 长腿黑丝高跟| 亚洲久久久久久中文字幕| 日韩人妻高清精品专区| 免费人成视频x8x8入口观看| 国产欧美日韩精品一区二区| avwww免费| 久久精品夜色国产| 亚洲成人av在线免费| 18禁在线播放成人免费| 日本一本二区三区精品| 美女免费视频网站| 日韩 亚洲 欧美在线| 精品不卡国产一区二区三区| 中国国产av一级| 亚洲人成网站高清观看| 3wmmmm亚洲av在线观看| 一进一出抽搐动态| 国产亚洲精品久久久久久毛片| 在线观看美女被高潮喷水网站| 国产色婷婷99| 久久99热这里只有精品18| 韩国av在线不卡| 国内精品美女久久久久久| 免费人成视频x8x8入口观看| 久久久久久久久中文| 99热只有精品国产| 中文字幕免费在线视频6| 一进一出好大好爽视频| 中文字幕久久专区| 国产蜜桃级精品一区二区三区| 日本黄色片子视频| 国产高潮美女av| 性色avwww在线观看| 大香蕉久久网| 免费大片18禁| 免费一级毛片在线播放高清视频| 日韩欧美一区二区三区在线观看| 国产单亲对白刺激| 人人妻,人人澡人人爽秒播|