• <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 ?

    Memristor-Based Genetic Algorithm for lmage Restoration

    2022-07-08 01:38:38YongBinYuChenZhouQuanXinDengYuanJingYangZhongManChengZhengFeiKang

    Yong-Bin Yu | Chen Zhou | Quan-Xin Deng | Yuan-Jing-Yang Zhong | Man Cheng | Zheng-Fei Kang

    Abstract—This paper explores a way of deploying the classical algorithm named genetic algorithm (GA) with the memristor.The memristor is a type of circuit device with both characteristics of storage and computing, which provides the similarity between electronic devices and biological components, such as neurons, and the structure of the memristor-based array is similar to that of chromosomes in genetics.Besides, it provides the similarity to the image gray-value matrix that can be applied to image restoration with GA.Thus, memristor-based GA is proposed and the experiment about image restoration using memristor-based GA is carried out in this paper.And parameters,such as the size of initial population and the number of iterations, are also set different values in the experiment,which demonstrates the feasibility of implementing GA with memristors.

    1.lntroduction

    Memristor, the abbreviation name of the memory resistor, was first proposed by Chua in 1971[1].From the point of view of the circuit theory, there should also be a circuit element to represent the relationship between the flux-linkage and current, then, the memristor was proposed.However, limited by the production technology at that time, it was not born until 2008 when the HP lab made the first memristor[2], arousing the research upsurge of memristors globally.Due to the special switching mechanism, memory, continuous input and output, and nano scale, memristors have great potential applications.For example, the memristorcomplementary metal-oxide-semiconductor transistor (memristor-CMOS) hybrid circuits could be used to realize neural networks[3].The memristor is also well-behaved in signal processing[4]and cyber-security[5].Recently, its value in image processing has been found with the application in image processing.The memristor bridge-based lowpass filter is designed along with an adaptive Gaussian filter, which is presented for denosing images[6]and has been applied to edge detection[7].Besides, it has good practical value in hand gesture recognition[8],[9].

    The genetic algorithm (GA) is a method that Holland[10]draws lessons from the basic principles of genetics to simulate natural evolution.It is used in designing the artificial adaptive system.A computational model is proposed based on natural selection that simulates Darwin’s theory of biological evolution.The basic idea of GA is to select individuals from the initial population, adopt the natural law of “survival of the fittest”, and then produce a new generation of population through hybridization and mutation, so as to evolve generation by generation until the goal is met.Nowadays, it works well in solving complex nonlinear combinatorial optimization problems, such as the traveling salesman problem[11]and image processing[12].

    In recent years, image restoration has become important in many fields, such as medical imaging and graphic printing.Digital images are often corrupted by degradation[13].The reason of image blur is complicated, the restoration cannot be processed by the regular linear function.So, it is proper to use GA to restore the image.

    Because of the advantage of memristors in storage, the memristor-based array is proposed, which also can be called the memristive network.This array is capable of solving more complex problems compared with the single memristor unit.Besides, it has the similarity to the image gray-value matrix, and its structure is similar to chromosomes and genes.In the memristor-based array, every pixel of the image has a corresponding memristor unit to store its value, making it possible to realize storage for binary, grayscale, and color images.So our goals are to verify the feasibility of deploying GA with memristors, explore mapping rules between GA and the memristor-based array, and deploy GA with memristors through the mapping rules in image restoration.

    2.Computing Model

    2.1.GA

    GA can be divided into three parts.1) The first part is initialization.Engineering problems are mapped into population evolution problems in the biological world.Coding and generating population and feasible solutions of problems can be defined by chromosomes and genes.2) The second part is to design the fitness function, that is, choosing the appropriate standard to quantify the individual’s adaptation to the living environment.Fitness is an important standard to evaluate the individual and the basis of other operations.Because the requirement of the fitness function is not particularly strict and it is not forced to be differentiable,derivable, and so on, GA is suitable to deal with nonlinear problems.3) The third part is the core operations of GA, including selection, crossover, and mutation.Selection is to select individuals with good characters in the current population, which is the basis of crossover and mutation operations.Crossover is the exchange of the segment of parents’ genes, that is, the process of reproduction in biology, which is an important part of GA.And the genes of parents can be continued because of crossover.Mutation is the idea of mutation in biology,sometimes, gene mutation occurs during the process that offspring inherits the parent gene, and mutation is beneficial for the population to jump out of the local optimal solution.The specific process of GA is showed inFig.1.

    2.2.Memristor-Based Array

    Limited by the computing ability of a single memristor, the memristor-based array was designed for more complex problems.The memristor-based array has the structure similarity to a two-dimensional matrix, which is composed of multiple memristors.Its structure is shown inFig.2.Each memristor can be added to this array through two connecting wires, and the deletion of the memristor only needs to disconnect the two connecting wires between the memristor and array[14].

    Fig.1.Process of GA.

    Fig.2.Structure of memristor-based array.

    There have been applications of using the memristive network to implement other swarm intelligence algorithms.For example, Pershin and Ventra[15]compared the memory characteristics of memristors with the pheromones left by ants on the path, proposed a mapping relationship, and used memristors to deploy the ant colony optimization, which inspires us to explore memristor-based GA.

    2.3.Memristor-Based GA

    To solve optimization problems, the analogy and mapping between GA and the memristor array is then conducted.The structure of the memristor-based array is similar to that of the chromosome, so a single memristor in the array can be abstracted into a gene unit.Then the operation in GA is mapped to the change of the resistance value of the memristor in the memristor-based array.The mapping relationship is shown inTable 1.The memristor is the basic unit of the memristor network and the gene is the basic unit of the chromosome; the sum of information of each pixel constitutes an image, so the memristor is mapped to genes in GA, which corresponds to the gray value information of a pixel in an image.Multiple genes form a chromosome or individual, so the chromosome corresponds to a set of memristors, that is, a memristor network and a set of gray value information of pixels in the memristor network.

    Table 1: Mapping rules among nature, GA, memristive network, and image restoration

    3.lmage Restoration Based on Memristor-Based GA

    This section will introduce the theory of image restoration, detailed design of image restoration based on GA, including the way of coding and generating initial population, the selection of the fitness function, and the selection of the basic operation of GA.

    3.1.Theory of lmage Restoration

    In the process of image generation, transportation, and processing, the phenomenon of image blur caused by man-made and natural factors, such as the camera jitter, inappropriate light, and haze, is called image degradation.Except image blur, the degradation is often manifested as distortion, additional noise, etc.,resulting in the decline in image quality.In the image degradation model, image degradation is modeled by using the degradation function and adding the external noise.The model is showed inFig.3,xandydenote coordinates in the spatial domain,f(x,y)represents the original clear image in the spatial domain,h(x,y) is the degradation function,n(x,y)represents the noise, andg(x,y) denotes the degraded image.

    The mathematical expression of the degenerate model can be expressed as[16]

    Fig.3.Image degradation model.

    where * denotes the convolution operation.After the Fourier transform, (1) in the frequency domain is

    It can be inferred that the restored images in the frequency domain is

    G(u,v) is defined in (2), and then substituted into (3):

    The restored image is obtained by using the inverse Fourier transform:

    wheref′(x,y) denotes the restored image.So, it is possible to restore a blurred image without nois e.

    Before the start of GA, the significance and design of image segmentation will be introduced.During the process of image restoration, the number of pixel in a picture is very big, so the computing burden will be heavy.Suppose that the size of the image to be restored is 100×30 pixels, and the image is not divided into blocks.At this time, an image is an individual has 100×30 = 3000 genes, and the range of each gene is an integer of 0 to 255.For one graph, there are 2563000different feasible solutions, which is a very big number.Usually, the size of the initial population is about 20 to 120.In this case, the size of the randomly generated initial population is only about 100/2563000, and the result of this ratio is close to zero.Therefore, the probability of getting individuals with good characters is very small, that is, the fitness of individuals is very low, and the use of improved selection, crossover, and mutation algorithms cannot get good results afterwards.And the speed of convergence is very slow, because the individual does not perform well from the beginning.Furthermore, in order to increase the probability of producing good initial individuals, we have to choose to increase the size of the initial population.However, increasing the size of the initial population means that more space is needed to store individual information, and each additional individual of the initial population needs more 3000 memories to store individual information, so the space cost is very huge.The most critical problem is that even if the size of the initial population is increased to thousands, tens of thousands, or even hundreds of thousands, it is still a very small number compared with 2563000.Moreover,when the size of the picture increases slightly, this number will grow exponentially, and the increasing population size will greatly increase the amount of memristors needed and the space cost.The effect on improving the efficiency of the algorithm is negligible.So, above all, we put forward to process the image by blocks.

    3.2.Divide lmage into Small Blocks

    In this situation, the size of the image block is set as 3×3 pixels, and the image block is moved according to the rules from left to right and from top to bottom.The 3×3-image block first traverses from left to right,moves one pixel down after traversing a whole line, and continues to traverse a whole row until all pixels have been traversed.

    Thus, an individual is not a whole image, but the size of a block, and its structure is like that a chromosome has nine genes.To simplify the operation, the data matrix of a 3×3 block will be converted into a matrix of 1×9.Assuming that the number of the initial population is 20, then the size of the initial population is the two-dimensional matrix of 20×9.Besides, each individual also needs a memory to store the value of fitness, so the size of the matrix is extended to 20×10.This matrix is named as the operation matrix, in which the first nine genes in each row form an individual.The number of rows represents the size of the population,and the last one stores the individual fitness value.So two memristor-based arrays are required, one is to store the blurred image (called MNblur instead), and another is to store the restored image (called MNres instead).In addition, an operation-matrix memristor-based array is also needed, GA operations all happen in the operation-matrix memristor-based array.

    3.3.Select GA Operation for lmage Restoration

    3.3.1.Coding and Initializing Population

    Two-dimensional integer coding is adopted by this paper, because the picture can be seen as a twodimensional matrix storing pixel information.The gray value of the image is expressed as the genetic value directly, which not only retains the similarity between the data and the image, but also avoids the complex coding and decoding operations.It is helpful to improve the operation speed, and then stochastically generate initial population.

    3.3.2.Fitness Function

    The fitness function is a function to evaluate the fitness of an individual, which is generally required to be non-negative.The purpose of image restoration is to obtain the best original image, but the restoration quality of an image can only be known by comparing with the original image and restored image, especially, the original image is unknown.However, we discussed the image degradation model in subsection 3.1, and the degraded imageg(x,y) is degraded from the original imagef(x,y). If the restored imagef′(x,y) can also produce another degraded imageg′(x,y) after the same degradation function, then bothg(x,y) andg′(x,y)are known and can be compared.If the finalg′(x,y) is closer tog(x,y),f′(x,y) will be more similar tof(x,y),so the fitness function is designed by[8]

    3.3.3.Selection Function

    The selection function in this research adopts the roulette selection: The probability of each individual being selected is defined as

    wherexirepresents individual, andf(xi) denotes the fitness value of individualxi.

    3.3.4.Crossover Function

    As we discussed before, we designed image block processing, and one image block is an individual.The number of genes is very small, so instead of the complex way of crossover, one-point crossover is used.The crossing point is selected by a random function.According to the position of the selected crossing point, a chromosome is divided into two parts before and after the crossing point, where the matching parents exchange the front or back gene values of the crossing point.

    3.3.5.Mutation Function

    The mutation function imitates the mutation of biological genes in nature.First, a gene bit on the chromosome is selected randomly, and the decimal gray value on the gene position is converted into the binary code of 8 bits, then one bit is randomly selected to change, and finally the changed binary value is converted into the decimal gray value.

    4.Simulation Results

    4.1.Data of Experiment

    The experimental data is the captured picture shown inFig.4, and the size is 101×26 pixels.The degraded image is obtained from the original image by a 4×4-point spread function, as shown inFig.5.

    Fig.4.Test image.

    Fig.5.Degraded image.

    4.2.Experiment Results

    We discussed the design of dividing image into blocks, so in Experiment 1, the results of dividing the image into blocks or not are compared.The set of parameters is shown inTable 2and the results are shown inFig.6.

    Figs.6(a) and (b) are the image restoration results of Case 1 and Case 2, designed inTable 2,respectively.They are set with the same value of the initial population and iteration, and the only difference is the operation of dividing image into blocks or not.Obviously, as shown inFig.6(b), by dividing the image into blocks, Case 2 has a better restoration result.Because of the large number of pixels and the wide range of pixels in the image processed without blocks, it is difficult to get good individuals by randomly generating the initial population.As a result, the convergence speed is very slow in the later stage, and the individual performance is still very poor after the same number of iterations.And it is less likely to obtain more information through the restored image in Case 1.

    Then in Experiment 2, different values of the population size and iterations are set inTable 3.For all cases inTable 3, the dividing operation with the block size of 3×3 pixels is adopted.

    Table 2: Case design

    Table 3: Different parameter values

    Fig.6.Image restoration results of case design in Table 2:(a) result of Case 1 and (b) result of Case 2.

    Then, we get different results and their running time, as shown inFig.7.

    Fig.7.Experiment 2 results: (a) ID = 2 and t = 50 min, (b) ID = 3 and t = 22 min, (c) ID = 4 and t = 10.5 min, and (d) ID = 5 and t = 7 min.

    4.3.Comparative Study

    The contrastive study is shown in this subsection inTable 4.And experiment results are shown inFig.8.

    Fig.8.Comparative study results: (a) Gaussian filter, (b) median filter, and (c) our proposed model.

    Figs.8(a) and (b) are the results of applying the Gaussian and median filters, respectively, andFig.8(c) is the results by applying the proposed model, which is better than the other two.From bothTable 4andFig.8, it can be seen that image restoration results are not good by directly performing the filtering operation on the image, and our proposed model performs better, which verifies the improvement of our model.

    Table 4: Compare the proposed method with filters

    Compared with Case 4, Case 3 has a larger number of iterations, leading to longer running time and a better result.Compared with Case 5, Case 4 has a larger population size, leading to longer running time and a clearer result.Cases 1 to 5 all demonstrate the feasibility of deploying GA with memristors in image restoration.Besides, when the initial population and the number of iterations are too small (as shown inFig.7(d)), the restoration result of the image is the worst: The words and numbers are vague.After increasing the size of the initial population and the number of iterations, the restoration results are improved,but it is not that: The larger the size of population and the number of iterations are, the better the results.Therefore, the appropriate population size and iteration time should be chosen according to the actual needs,without blindly pursuing the large size of the initial population and iterations.Although the restoration result could be improved, the time cost is very large.So both the improvement of restoration and time cost should be taken into consideration.

    5.Conclusions

    In this paper, memristor-based GA for image restoration is proposed.The mapping rules between the memristor and GA are shown.According to the rules, memristor-based GA is designed and carried out successfully.The comparative study shows that the proposed method has better performance.It is demonstrated that using memristors to implement GA is possible and memristor-based GA for image restoration is feasible.

    Memristor, GA, and image restoration all have wide applications in different fields.There are still some other algorithms, like the ant colony optimization, which can be implemented in a similar way.But the existing methods are not completely unified yet.Once unified, more different algorithms can be implemented with the memristor-based array, which is helpful to increase the use of memristors.So more efforts are needed to design a unified system and enrich the application cases.

    Disclosures

    The authors declare no conflicts of interest.

    国产单亲对白刺激| 又爽又黄a免费视频| 有码 亚洲区| 男女做爰动态图高潮gif福利片| 国产高清视频在线播放一区| 亚洲精品在线观看二区| 国产乱人视频| 亚洲国产精品sss在线观看| 一区福利在线观看| 有码 亚洲区| 久久久久久大精品| 夜夜爽天天搞| 欧美黑人欧美精品刺激| 成人国产一区最新在线观看| 国产伦精品一区二区三区视频9| 亚洲av二区三区四区| 亚洲中文字幕一区二区三区有码在线看| 国产高清视频在线播放一区| 午夜a级毛片| 亚洲精品粉嫩美女一区| 一个人观看的视频www高清免费观看| 久久6这里有精品| 春色校园在线视频观看| 99九九线精品视频在线观看视频| 国产一区二区亚洲精品在线观看| 国产av不卡久久| a级毛片a级免费在线| or卡值多少钱| 国产一区二区三区在线臀色熟女| 久久这里只有精品中国| 高清日韩中文字幕在线| 精品午夜福利在线看| or卡值多少钱| 啪啪无遮挡十八禁网站| 高清毛片免费观看视频网站| 一级黄片播放器| 91精品国产九色| 午夜久久久久精精品| 久久精品久久久久久噜噜老黄 | av黄色大香蕉| 国产av一区在线观看免费| 一夜夜www| 嫩草影院精品99| 偷拍熟女少妇极品色| 色播亚洲综合网| 麻豆成人午夜福利视频| 美女 人体艺术 gogo| 性欧美人与动物交配| 免费黄网站久久成人精品| av在线老鸭窝| 国产在线精品亚洲第一网站| 国产av麻豆久久久久久久| 国产探花在线观看一区二区| 国产精品亚洲一级av第二区| 一a级毛片在线观看| 国国产精品蜜臀av免费| 禁无遮挡网站| 色综合婷婷激情| 波多野结衣高清无吗| 久久久久久大精品| 亚洲av不卡在线观看| 伊人久久精品亚洲午夜| 国产真实伦视频高清在线观看 | 欧美最新免费一区二区三区| 亚洲熟妇中文字幕五十中出| 久久国内精品自在自线图片| 久久久久久久亚洲中文字幕| 男人舔奶头视频| 久久久精品欧美日韩精品| 国产精品美女特级片免费视频播放器| 亚洲avbb在线观看| 亚洲午夜理论影院| 亚洲成人久久性| 日韩大尺度精品在线看网址| 久久久精品大字幕| 久久久精品大字幕| 国产精品一区二区三区四区免费观看 | 精品人妻视频免费看| 亚洲乱码一区二区免费版| eeuss影院久久| 亚洲美女视频黄频| 久久中文看片网| 哪里可以看免费的av片| 精品乱码久久久久久99久播| a级毛片免费高清观看在线播放| 免费看美女性在线毛片视频| 亚洲黑人精品在线| www.www免费av| 99热这里只有是精品在线观看| 99热这里只有精品一区| 国产伦精品一区二区三区四那| 免费在线观看日本一区| 白带黄色成豆腐渣| 最好的美女福利视频网| 国产毛片a区久久久久| 亚洲精品久久国产高清桃花| 一本久久中文字幕| 97超级碰碰碰精品色视频在线观看| 在线观看午夜福利视频| 午夜爱爱视频在线播放| x7x7x7水蜜桃| 婷婷丁香在线五月| 99热这里只有是精品50| 日韩在线高清观看一区二区三区 | 免费在线观看影片大全网站| 国产一区二区三区在线臀色熟女| 国产免费av片在线观看野外av| 免费高清视频大片| 极品教师在线免费播放| 韩国av在线不卡| 精品国内亚洲2022精品成人| 亚洲一区二区三区色噜噜| 午夜a级毛片| 男女边吃奶边做爰视频| 久久精品国产亚洲av香蕉五月| 欧美激情在线99| 亚洲人成伊人成综合网2020| 精品国内亚洲2022精品成人| av天堂中文字幕网| 搡老岳熟女国产| 国国产精品蜜臀av免费| 国内精品宾馆在线| 精品99又大又爽又粗少妇毛片 | 91av网一区二区| 国产伦人伦偷精品视频| 亚洲精品456在线播放app | 男女之事视频高清在线观看| 久久午夜亚洲精品久久| 亚洲欧美精品综合久久99| 久久精品国产亚洲av涩爱 | 美女高潮的动态| 国产欧美日韩精品一区二区| 不卡一级毛片| 天堂√8在线中文| 精品一区二区三区视频在线| 99久久成人亚洲精品观看| 熟女电影av网| 深夜精品福利| 真人一进一出gif抽搐免费| 天堂网av新在线| 国产伦精品一区二区三区四那| 成年版毛片免费区| 午夜福利欧美成人| 天天躁日日操中文字幕| 婷婷丁香在线五月| 日韩大尺度精品在线看网址| 日韩,欧美,国产一区二区三区 | 真人一进一出gif抽搐免费| 国产欧美日韩精品一区二区| 日韩欧美国产一区二区入口| 在线观看av片永久免费下载| 亚洲精品国产成人久久av| 亚洲五月天丁香| 欧美中文日本在线观看视频| 欧美潮喷喷水| 变态另类成人亚洲欧美熟女| av天堂中文字幕网| av国产免费在线观看| a级毛片免费高清观看在线播放| 亚洲av免费高清在线观看| 99热精品在线国产| 亚洲av中文av极速乱 | 国产成人一区二区在线| 色吧在线观看| 国产精品伦人一区二区| 在线天堂最新版资源| 成人特级av手机在线观看| 国产精品自产拍在线观看55亚洲| 高清毛片免费观看视频网站| 欧美+日韩+精品| 黄色女人牲交| 麻豆精品久久久久久蜜桃| 99久久九九国产精品国产免费| 丰满人妻一区二区三区视频av| 久久午夜亚洲精品久久| 色综合色国产| 亚洲av成人av| 中出人妻视频一区二区| 成人高潮视频无遮挡免费网站| 免费电影在线观看免费观看| 国产精品伦人一区二区| 久久热精品热| 久久久久九九精品影院| 一个人看视频在线观看www免费| 丰满的人妻完整版| 国产69精品久久久久777片| av天堂中文字幕网| 国产黄色小视频在线观看| 国产综合懂色| 免费看a级黄色片| 精品久久久久久久末码| 日日摸夜夜添夜夜添av毛片 | 国产精品野战在线观看| 99久久精品热视频| 久久草成人影院| 在线天堂最新版资源| 中文字幕熟女人妻在线| 国产色婷婷99| 男女做爰动态图高潮gif福利片| 成人av在线播放网站| 伊人久久精品亚洲午夜| 变态另类丝袜制服| 99热这里只有精品一区| 午夜福利18| 俺也久久电影网| netflix在线观看网站| 色综合亚洲欧美另类图片| 欧美xxxx性猛交bbbb| 日韩精品中文字幕看吧| 日韩强制内射视频| 婷婷丁香在线五月| 欧美最黄视频在线播放免费| 亚洲va日本ⅴa欧美va伊人久久| 精品人妻1区二区| 亚洲av熟女| 日本三级黄在线观看| 久久久久久九九精品二区国产| 搡老妇女老女人老熟妇| 久久精品国产亚洲av天美| 一级黄色大片毛片| 99久久中文字幕三级久久日本| 免费人成视频x8x8入口观看| 中出人妻视频一区二区| 国产精品乱码一区二三区的特点| 少妇被粗大猛烈的视频| 欧美黑人巨大hd| 亚洲国产精品久久男人天堂| av在线亚洲专区| av在线观看视频网站免费| 欧美一区二区亚洲| aaaaa片日本免费| 少妇的逼水好多| 中文字幕av成人在线电影| 亚洲av日韩精品久久久久久密| 熟女人妻精品中文字幕| 成人欧美大片| 22中文网久久字幕| 九色成人免费人妻av| 日韩亚洲欧美综合| 男女那种视频在线观看| 精品一区二区三区人妻视频| 国产亚洲91精品色在线| 1024手机看黄色片| 中文字幕久久专区| 日本免费一区二区三区高清不卡| 色综合站精品国产| 性欧美人与动物交配| 伦理电影大哥的女人| 99久久精品热视频| 男人和女人高潮做爰伦理| 真人做人爱边吃奶动态| 草草在线视频免费看| 日本 欧美在线| 亚洲av成人精品一区久久| 婷婷精品国产亚洲av在线| 亚洲七黄色美女视频| 天堂影院成人在线观看| 国内精品美女久久久久久| АⅤ资源中文在线天堂| 国产淫片久久久久久久久| 国产在线精品亚洲第一网站| 国语自产精品视频在线第100页| 草草在线视频免费看| 久久久久国内视频| 免费观看精品视频网站| 在线观看av片永久免费下载| 免费高清视频大片| 国产精品一区二区性色av| АⅤ资源中文在线天堂| 一级黄片播放器| 老熟妇乱子伦视频在线观看| 欧美黑人欧美精品刺激| 非洲黑人性xxxx精品又粗又长| 亚洲va日本ⅴa欧美va伊人久久| 亚洲自拍偷在线| 岛国在线免费视频观看| 又爽又黄无遮挡网站| 免费看光身美女| 国产精品国产三级国产av玫瑰| 精品一区二区三区视频在线| 精品久久久噜噜| 搞女人的毛片| 日本熟妇午夜| 又紧又爽又黄一区二区| 婷婷精品国产亚洲av在线| 俺也久久电影网| av在线亚洲专区| 精品人妻偷拍中文字幕| 天堂动漫精品| 夜夜夜夜夜久久久久| 亚洲人成网站高清观看| bbb黄色大片| 特级一级黄色大片| 色噜噜av男人的天堂激情| av视频在线观看入口| 亚洲精品乱码久久久v下载方式| 日本成人三级电影网站| 97超视频在线观看视频| 久久久色成人| 欧美绝顶高潮抽搐喷水| 国产主播在线观看一区二区| 久久亚洲真实| 午夜久久久久精精品| 久久欧美精品欧美久久欧美| 男女之事视频高清在线观看| 国产精品永久免费网站| 国产极品精品免费视频能看的| 久久久久久国产a免费观看| 国产精品伦人一区二区| 国产白丝娇喘喷水9色精品| 国产精品,欧美在线| 热99re8久久精品国产| 麻豆成人av在线观看| 欧美日韩乱码在线| 欧美潮喷喷水| 国产精品久久久久久精品电影| 亚洲欧美清纯卡通| 午夜免费男女啪啪视频观看 | 亚洲一区二区三区色噜噜| 俄罗斯特黄特色一大片| 成人av在线播放网站| 久久国内精品自在自线图片| 欧美日本视频| 88av欧美| 久久精品国产亚洲av天美| 国产精品亚洲一级av第二区| 欧美一区二区亚洲| 99精品在免费线老司机午夜| 免费av观看视频| 神马国产精品三级电影在线观看| 亚洲精品影视一区二区三区av| 日日摸夜夜添夜夜添av毛片 | 国内精品久久久久久久电影| 999久久久精品免费观看国产| 亚洲七黄色美女视频| 美女被艹到高潮喷水动态| 97碰自拍视频| 亚洲av免费高清在线观看| 欧美绝顶高潮抽搐喷水| 精品午夜福利在线看| 亚洲不卡免费看| 色哟哟哟哟哟哟| 99久久无色码亚洲精品果冻| 啦啦啦韩国在线观看视频| 一区福利在线观看| 麻豆国产av国片精品| 真人做人爱边吃奶动态| 亚洲美女黄片视频| 亚洲乱码一区二区免费版| 岛国在线免费视频观看| 欧美一级a爱片免费观看看| 中文字幕av成人在线电影| 精品一区二区三区视频在线| 精品人妻偷拍中文字幕| 亚洲欧美激情综合另类| 最好的美女福利视频网| 啦啦啦啦在线视频资源| av国产免费在线观看| 老司机午夜福利在线观看视频| 亚洲欧美日韩无卡精品| 国产69精品久久久久777片| 2021天堂中文幕一二区在线观| 在线免费十八禁| 变态另类成人亚洲欧美熟女| 俄罗斯特黄特色一大片| 久久久久精品国产欧美久久久| 无遮挡黄片免费观看| 日韩亚洲欧美综合| 国产精品98久久久久久宅男小说| 亚洲av一区综合| 麻豆一二三区av精品| 国产成人福利小说| 人人妻人人看人人澡| 91久久精品国产一区二区三区| 中文字幕av在线有码专区| 噜噜噜噜噜久久久久久91| 看片在线看免费视频| 丰满乱子伦码专区| netflix在线观看网站| 嫩草影院精品99| 亚洲精华国产精华液的使用体验 | 亚洲真实伦在线观看| x7x7x7水蜜桃| 97人妻精品一区二区三区麻豆| 免费看av在线观看网站| avwww免费| 国产精品永久免费网站| 国产成年人精品一区二区| 成人美女网站在线观看视频| 九色成人免费人妻av| 天堂av国产一区二区熟女人妻| 亚洲无线在线观看| 熟妇人妻久久中文字幕3abv| 女人十人毛片免费观看3o分钟| 亚洲av五月六月丁香网| 俺也久久电影网| 不卡一级毛片| 精品无人区乱码1区二区| 国产精品久久久久久亚洲av鲁大| aaaaa片日本免费| 国产午夜精品久久久久久一区二区三区 | 两个人视频免费观看高清| 亚洲av五月六月丁香网| 日本一二三区视频观看| 我的老师免费观看完整版| 亚洲国产色片| 精品人妻1区二区| 又爽又黄无遮挡网站| 欧美bdsm另类| 精品久久久久久久久久久久久| 美女黄网站色视频| 日韩一区二区视频免费看| 久久久精品欧美日韩精品| 色5月婷婷丁香| 欧美日韩乱码在线| 在线观看66精品国产| 给我免费播放毛片高清在线观看| ponron亚洲| 99热6这里只有精品| 日本与韩国留学比较| 亚洲精品456在线播放app | 国产探花在线观看一区二区| 国产伦精品一区二区三区视频9| 国产亚洲欧美98| 国产欧美日韩精品一区二区| 免费在线观看影片大全网站| 日本欧美国产在线视频| bbb黄色大片| 精品不卡国产一区二区三区| 成人午夜高清在线视频| 韩国av一区二区三区四区| 成人特级av手机在线观看| 大又大粗又爽又黄少妇毛片口| 久久久久久伊人网av| 成年女人毛片免费观看观看9| 久久热精品热| 久久久久九九精品影院| 伦理电影大哥的女人| 俄罗斯特黄特色一大片| 国产精品电影一区二区三区| 他把我摸到了高潮在线观看| 久久99热6这里只有精品| 国产高清视频在线播放一区| 久久午夜福利片| 尾随美女入室| 少妇的逼好多水| 久久久久精品国产欧美久久久| 深夜a级毛片| 日韩亚洲欧美综合| 亚洲专区国产一区二区| 18禁在线播放成人免费| 天堂网av新在线| 丰满的人妻完整版| 亚洲一区高清亚洲精品| 亚洲 国产 在线| 99久久精品国产国产毛片| 国产乱人伦免费视频| 国产精品女同一区二区软件 | 日本与韩国留学比较| 精品一区二区三区视频在线| 久久精品91蜜桃| 午夜福利在线观看吧| 欧美激情久久久久久爽电影| 午夜亚洲福利在线播放| 91狼人影院| 国产欧美日韩一区二区精品| 亚洲成人久久性| 国产中年淑女户外野战色| 亚洲乱码一区二区免费版| 精品日产1卡2卡| 91麻豆精品激情在线观看国产| 国产精品野战在线观看| 成人午夜高清在线视频| 国产午夜福利久久久久久| 很黄的视频免费| 五月伊人婷婷丁香| 精品一区二区三区av网在线观看| 俄罗斯特黄特色一大片| 国产老妇女一区| 精品人妻熟女av久视频| 亚洲国产精品sss在线观看| 日韩欧美国产在线观看| 综合色av麻豆| 国产黄片美女视频| 麻豆成人午夜福利视频| 18禁黄网站禁片免费观看直播| 乱码一卡2卡4卡精品| 亚洲无线在线观看| 亚洲色图av天堂| 最近在线观看免费完整版| 亚洲真实伦在线观看| 日本在线视频免费播放| 欧美日韩中文字幕国产精品一区二区三区| 黄色一级大片看看| 在现免费观看毛片| 国产真实伦视频高清在线观看 | 91久久精品国产一区二区三区| 黄色一级大片看看| 麻豆成人午夜福利视频| av国产免费在线观看| 国产精品一区二区免费欧美| 国产精品女同一区二区软件 | 午夜激情福利司机影院| 午夜a级毛片| 欧美最黄视频在线播放免费| 国产高清有码在线观看视频| 国产一级毛片七仙女欲春2| 在线看三级毛片| 久久久成人免费电影| 国产在线男女| 色精品久久人妻99蜜桃| 日韩欧美国产一区二区入口| 久久精品国产鲁丝片午夜精品 | 亚洲av五月六月丁香网| 人妻少妇偷人精品九色| 最新中文字幕久久久久| 窝窝影院91人妻| 成人午夜高清在线视频| 99久久久亚洲精品蜜臀av| 男人狂女人下面高潮的视频| 日本精品一区二区三区蜜桃| 国产伦一二天堂av在线观看| 国产亚洲精品久久久com| 久久人人精品亚洲av| 老女人水多毛片| 欧美性感艳星| 国产熟女欧美一区二区| 国产69精品久久久久777片| 丰满人妻一区二区三区视频av| 亚洲精华国产精华液的使用体验 | 日本爱情动作片www.在线观看 | 亚洲av美国av| 免费在线观看日本一区| 久久精品综合一区二区三区| 人人妻人人看人人澡| 超碰av人人做人人爽久久| 黄色配什么色好看| 一边摸一边抽搐一进一小说| 12—13女人毛片做爰片一| 免费看日本二区| 国产男人的电影天堂91| 不卡视频在线观看欧美| 在现免费观看毛片| www日本黄色视频网| 中文字幕久久专区| 91久久精品国产一区二区成人| 悠悠久久av| 亚洲男人的天堂狠狠| 日韩欧美一区二区三区在线观看| 国产在视频线在精品| 五月玫瑰六月丁香| 色综合婷婷激情| 性色avwww在线观看| 国产三级在线视频| 成人美女网站在线观看视频| 一区二区三区四区激情视频 | 女同久久另类99精品国产91| 美女高潮的动态| 亚洲自偷自拍三级| 日韩精品中文字幕看吧| 51国产日韩欧美| 国产伦精品一区二区三区视频9| 午夜免费男女啪啪视频观看 | 精品一区二区三区av网在线观看| 一区二区三区四区激情视频 | 午夜福利在线观看免费完整高清在 | 中文字幕熟女人妻在线| 一区二区三区激情视频| 蜜桃亚洲精品一区二区三区| 有码 亚洲区| av天堂中文字幕网| 啦啦啦啦在线视频资源| 长腿黑丝高跟| 少妇高潮的动态图| 国内精品久久久久久久电影| 亚州av有码| 日韩精品中文字幕看吧| 在线免费观看不下载黄p国产 | 亚洲成av人片在线播放无| 亚洲欧美清纯卡通| 国产 一区精品| 窝窝影院91人妻| videossex国产| 亚洲成人精品中文字幕电影| 天堂网av新在线| 毛片一级片免费看久久久久 | 午夜爱爱视频在线播放| 精品一区二区免费观看| 国产精品99久久久久久久久| 好男人在线观看高清免费视频| 久久精品久久久久久噜噜老黄 | 97热精品久久久久久| 亚洲自拍偷在线| 可以在线观看毛片的网站| 久久久久国内视频| 97人妻精品一区二区三区麻豆| 日韩一区二区视频免费看| 在线观看午夜福利视频| 日本色播在线视频| 午夜精品一区二区三区免费看| 一个人观看的视频www高清免费观看| 亚洲 国产 在线| 男女那种视频在线观看| 日韩av在线大香蕉| 波多野结衣高清无吗| 精品久久久噜噜| 亚洲精品影视一区二区三区av| 村上凉子中文字幕在线| 日韩一区二区视频免费看| a在线观看视频网站| 日本色播在线视频| 国产高清三级在线| 色播亚洲综合网| 亚洲美女黄片视频| 免费一级毛片在线播放高清视频|