• <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 ?

    Hybrid GrabCut Hidden Markov Model for Segmentation

    2022-08-24 12:58:00SoobiaSaeedAfnizanfaizalAbdullahJhanjhiMehmoodNaqviMehediMasudandMohammedAlZain
    Computers Materials&Continua 2022年7期

    Soobia Saeed, Afnizanfaizal Abdullah, N.Z.Jhanjhi, Mehmood Naqvi, Mehedi Masudand Mohammed A.AlZain

    1Department of Software Engineering, UniversitiTeknologi Malaysia-UTM, 81310, Malaysia

    2School of Computer Science and Engineering, SCE Taylor’s University, Subang Jaya, 47500, Malaysia

    3Department of Engineering and Technology, Mohak College, Alberta, Canada

    4Department of Computer Science, College of Computers and Information Technology, Taif University,Taif, 21944, Saudi Arabia

    5Department of Information Technology, College of Computers and Information Technology, Taif University,Taif, 21944, Saudi Arabia

    Abstract: Diagnosing data or object detection in medical images is one of the important parts of image segmentation especially those data which is less effective to identify in MRI such as low-grade tumors or cerebral spinal fluid(CSF) leaks in the brain.The aim of the study is to address the problems associated with detecting the low-grade tumor and CSF in brain is difficult in magnetic resonance imaging (MRI) images and another problem also relates to efficiency and less execution time for segmentation of medical images.For tumor and CSF segmentation using trained light field database(LFD) datasets of MRI images.This research proposed the new framework of the hybrid k-Nearest Neighbors (k-NN) model that is a combination of hybridization of Graph Cut and Support Vector Machine (GCSVM) and Hidden Markov Model of k-Mean Clustering Algorithm (HMMkC).There are four different methods are used in this research namely (1) SVM, (2)GrabCut segmentation, (3) HMM, and (4) k-mean clustering algorithm.In this framework, on the one hand, phase one is to perform the classification of SVM and Graph Cut algorithm to create the maximum margin distance.This research use GrabCut segmentation method which is the application of the graph cut algorithm and extract the data with the help of scaleinvariant features transform.On the other hand, in phase two, segment the low-grade tumors and CSF using a method adapted for HMkC and extract the information of tumor or CSF fluid by GCHMkC including iterative conditional maximizing mode (ICMM) with identifying the range of distant.Comparative evaluation is also performing by the comparison of existing techniques in this research.In conclusion, our proposed model gives better results than existing.This proposed model helps to common man and doctor that can identify their condition of brain easily.In future, this will model will use for other brain related diseases.

    Keywords: SVM; tumor; CSF; k-NN; MRI; grabcut segmentation; HMM

    1 Introduction

    The concept of“K-Nearest Neighbor (k-NN)”is one of the simplest, easy-to-understand, and interprets non-parametric machine learning algorithms.It gives relatively high and competitive results.It is a multipurpose algorithm and can be used to figure out the classification as well as regression issues.New records are identified by k-NN by combining most historical records of“k”[1,2].For the last four decades, it is the widely used as a statistical method in pattern recognition because of its simplicity and easiness.It is a very popular algorithm for text categorization and is used in various research strategies.To implement the k-NN algorithm we have to determine the following:

    i.The distance between the new and the training samples,

    ii.The nearest K neighbors,

    iii.The classification of the new sample,

    If all samples possess a similar category, then every new sample will be grouped in this category or else, a new category is to be determined for each new sample following certain rules.It is assumed that neighboring data points feature similarity that is why the nearest neighbor method is used for prediction.Fundamentally, the nearest neighbor algorithm concept defines the“k”points which are nearest to the unidentified sample in the multidimensional space, sorts them, and assigns a class based on how closely it matches the k points [3].These k points are labeled as the k-nearest neighbors of the unknown samples.It is assumed in this algorithm that in multidimensional space,every instance resembles other points.The Euclidean distance is used to evaluate the nearest neighbor of any instance along with recalculating the new group“k”points [4,5].To find the nearest distance,we repeat this strategy until the“k”point selection.In this research, we use three techniques that rendered a lot in solving the k-NN missing data problem.The most common technique in k-NN algorithm is the support vector machine (SVM).In machine learning (ML), kNN and SVM illustrate various crucial transactions [6,7].Even if the list of input variables, the SVM approach predicts exactly for all new observations.Hence, SVM is one instant and ready-to-use approach that many technicians use for brain imaging.Many inputs but small sample sizes will be used for two application domains.SVM is a frequently operating technique to fulfill the fundamentals of k-NN algorithm that enhances the gap or difference between the categories and leads to efficient overall performance.Comparability comparison is used for various categories in SVM.Therefore, to find accuracy and clarity in classification we have used SVM in this research.Here this research extends the method by using the HMM so that the tumor classification based on previous and subsequent segmentation results will be concluded.This approach uses probabilistic reasoning over time and space for brain tumor segmentation from 4D MRI [8-12].The diagnosis of brain tumor and CSF liquid usually requires image segmentation to evaluate tumor detection in brain images.Since the structural changes of the tumor interact with other normal tissues, separately segmenting each target and then combining them to build up whole-brain images would not be efficient for the three-dimensional (3D)segmentation method [13].On the other hand, it is assumed that multi-dimensional (4D) segmentation(4D)segmentation has been widely used in the image segmentation field and proven to be effective [13-15].

    This research aims to examine whether the CSF leakage fluid and low-grade tumor (initial stage of the brain tumor) images are helpful to improve the quality of images using the GrabCut algorithm of MRI images in terms of both accuracy and quality.The datasets used in this research is a series of chronological low quality MRI images of brain tumor (cancer) and CSF fluid leakage in patient’s brain images.In addition, the previous research contribution is based on the 3D segmentation method that tried to enhance the quality of images to gain a higher resolution and remove the noise or existing errors [16].Therefore, the traditional method (3D) is not enough to enhance the accuracy of poor images of fluid as CT myelography or MR Myelography is more significant than MRI in identifying the area of CSF leak as if it arises from a considerable tear [17-22].

    In addition, the problem associated with this research also reduces k-NN missing imputation in poor MRI images as the K-NN algorithm is susceptible to outliers because its selection criteria are based on distance neighbors for the closet point [23].Essentially, k-NN cannot address the problem of missing imputation.The k-NN algorithm tries to achieve optimal searching factors to address the missing values, whereas other algorithms fail [24-26].Generally, this research needs more consequences about the outcomes of hybrid k-NN algorithms, which can reflect the improvement in accuracy using the 4D images segmentation method [27-31].

    Many other researchers and scholars used hybrid k-NN algorithm for classification of brain tumor of both malignant and benign.The researcher used hybrid ensemble method of random forest(RF) with combination of k-NN and decision tree (KNNRF-DT) based on majority voting method.The researchers focus the area of tumor and classify the both type of tumor, in the starting section segmentation were performed using the Otsu’s threshold method.Feature extraction is performed by Stationary Wavelet Transform (SWT), Principle Component Analysis (PCA), and Gray Level Cooccurrence Matrix (GLCM), which provide the thirteen features for classification of hybrid k-NN ensemble classifier [32,33].One of the other researches is relevant to brain tumor which improve the performance of hybrid k-NN algorithm by the combination of SVM with k-NN for segmentation of brain tumor and increase the accuracy.The main contribution of this research is based on detecting the tumor region based on MRI was highlighted by the dim scale and symmetrical method and then applied hybrid k-NN algorithm to increase the performance of MRI and accuracy of the images [34-37].

    Our proposed approach is different than others as this research focus the low-grade tumor which is in sizable form and CSF fluid through hybrid k-NN model which is the combination of four different techniques.These techniques are not only maintained the accuracy and quality of images but increase the performance of proposed hybrid k-NN model.The combination these techniques namely GCHMkC which enhance the accuracy of the images that can tumor or fluid can easily identified in the images.The experimental results show the improve performance of this novel proposed model that enhance the quality and accuracy of this model.The structure and organization of this research work are detailed as follows: Section 1 interprets the research done previously, the current body of literature on this research’s main topics, and presents the related work.It also describes the writings on the application of 4D image of LFT segmentation and k-NN algorithm in different fields of research.Section2 presents our approach,methodology,and a brief review of the framework adopted.Section3 shows the experimental results and discussion of our proposed methods; Section 4 shows the proposed algorithm, and Section 5 show the conclusion and future work.

    2 Methods

    2.1 Overview

    The Proposed Hybrid Graph Cut Hidden Markov Model of k-mean Cluster (GCHMMkC)technique in the k-NN model that we have developed consists of two main components: Graph-Cut Support Vector Machine(GCSVM)technique and Hidden Markov Model of k-mean cluster(HMkC)approach as illustrated in Fig.1.The proposed techniques are based on multiple phases to implement in MRI images that are sufficient for the segmentation and classification of all types of images.The first component of the proposed model is based on the GCSVM technique, which consists of three phases:(1) Graph cut algorithm (2) GrabCut segmentation, and (3) SVM classification.These phases perform operations such as preparing the images for the classification of regression methods for training and labeling the datasets to enhance the accuracy of the images by the use of SVM.Graph cut algorithm transforms the data to generate the new elements of datasets.Furthermore, GrabCut segmentation’s results are link with the scale-invariantfeature transform (SIFT) function to identify the pixel size and range of fluid or tumor and connect to the second phase of the technique.The second component of our proposed model is the HMM and k-mean clustering algorithm (HMkC) technique, which consists of two phases: (1) HMM, and (2) k-mean clustering algorithm.These phases perform operations such as refining and reconstructing images using iterated conditional maximizing mode (ICMM) to join the probability of the trained dataset’s images with statistical confidence interval (d).The above two methods show the range of specified probabilities of HMM joined together, and statistical confidence intervals arrange the constructed images using ICMM to generate the same sequence form.In the end,the k-mean cluster is used to store the data for the classification of trained datasets.The explanation of the proposed technique is given below in detail in the next subsection.

    Figure 1: Foreground and background histogram models in a grayscale image

    2.2 Proposed Framework

    Quality is one of the most common issues in medical image datasets, which is subject to various factors such as noise in the images, irrelevant features, or missing boundaries.These issues need to be addressed before any segmentation method or classification of medical images is performed.Therefore, the proposed technique, explained next, reduces noise and irrelevant features in the images by the SVM and graph cut segmentation method and also apply a k-mean clustering algorithm to store the data for multi-classification of SVM.Firstly, SVM selects the images for training sample data and calculates the feature vectors of the trained datasets.SVM calculates feature vectors of the MRI image pixel point and gets them classified.In the second step, the previously selected sample points of the datasets from the target image are applied to the trained SVM to classify these image points.Furthermore, enhance the interested region points which are obtained in the second step to achieve the classification of an image and use the graph cut algorithm to help segment the tumor and CSF fluid in the images by using the segmentation technique of GrabCut segmentation method.In this research, the graph cut algorithm using the segmentation technique of GrabCut is combined with the SVM known as GCSVM hybridization.GCSVM increases the resolution of an image for training and labels the trained datasets.The image is treated as a graph structure with vertices and edges to define the structure of the image.Each pixel is represented by a vertex and neighboring pixels are linked by a weighted edge based on their similarity.In the case of multi-label image segmentation, each label has a unique vertex known as a terminal.The probability of the label assignment determines the vertices of pixels linked to all terminals and their edge weights.The solution to finding the minimum amount of energy lies in obtaining the segmentation that cuts on the graph at a minimum cost that a min-cut/maxflow algorithm can solve.Each vertex is connected after cutting the single terminal, which means the corresponding label is assigned to the corresponding pixel.This technique works on a combination of graph cut and SVM techniques.It maximizes the margin to become a max-margin graph cut for training the datasets in machine learning algorithms.This improves the hybrid GCSVM technique’s efficiency due to reducing the time and space complexity during the segmentation process for the SVM and graph cut algorithm to create the max-margin distance.

    2.3 Grabcut Segmentation

    GrabCut is a segmentation tech nique that uses graph cuts to perform segmentation for encapsulating the information in images.Most techniques use either edge detection or region detection, but GrabCut detects both of them.This information is used to create energy functions to be minimized and produce segments.To segment images, a graph is constructed with nodes representing pixels in the images.Additionally, two special nodes are created.These are known as“sink”and“source”nodes because they are linked to every pixel node in the graph.The sink node represents the image’s background, while the source node represents the image’s foreground.Both nodes must be separated from each other to segment the image.The energy function combines weights between pixel nodes as well as weights between pixel and source and sinks nodes in the graph.The weight between the pixel nodes determines the edge information in the image.These illustrate the edge between two pixels, such as a significant difference in pixel color resulting in a very small weight difference between two-pixel nodes.The weight between the Source and Sink nodes and pixel nodes are determined by the region information.These weights are determined by determining whether a pixel node is in the background or foreground region.

    In the GrabCut segmentation process, a few pixels are labeled before the segmentation as foreground or background, which is needed to create the regions.This study selects pixels that are set apart during the clue marking stage and uses them to display model region data by constructing foreground and background histograms.Histograms are generated by noticing the frequency with which a pixel appears within a group of pixels.The intensity of grayscale images can range between 0 and 255 for each pixel.For grayscale images, a histogram with 256 bins is created, with each bin containing the frequency at which pixels with the intensity of the bin number occur in an image.Fig.1 depicts the use of foreground and background information to create histograms in a grayscale image.The probability of the pixel node lies between the weight on it, and the source node determines the foreground histogram.Furthermore, the probability determined by the weight between the pixel node and the sink node lies inside the background histogram.

    Fig.1 shows that pixels with a high-intensity value have a higher probability of appearing in the background histogram than in the foreground histogram.The edge weight between the pixel node and the Sink node is stronger than the edge weight between the pixel node and the Source node.Pixels with a low-intensity value have a higher probability of appearing in the foreground histogram than in the background histogram, so the edge weight between the pixel node and the Source node will be more grounded than the edge weight between the pixel node and the Sink node.Both foreground and background histograms are utilized as a resulting set of inputs for MRI image segmentation.This research applied binary labels 446×2 and 48761×2 set to foreground and background.These binary labels values of both background and foreground construct the graph structure such as G = (vertex,ε), where the vertex is the set of pixels or vertices, and ε is the set of assembly edges for the closest four connected pixels to each other.To find the vertex label with a minimum energy function, this research uses maximum-flow and minimum-cut methods.In the min-max cut algorithm, each image region was labeled to pixel intensity using the binary transformation function and stores the values of sink and source nodes.

    The aim of this section is to evaluate the predictive analysis of MRI data for the classification of tumor or CSF fluid leak after utilizing the technique of hybrid GCSVM to solve the classification problem.The image is transformed and is assigned the new data elements after classification for labeling the categories in terms of a binary classifier to receive the optimized solution.This proposed technique fulfils the requirement of segmentation’s outcomes and enhances the quality of images.The hybrid GCSVM technique classifies between these two classes, which are constructed by a hyper plane in 4-dimensional features utilized by the classification.In the end,SVM transforms the medical images datasets by the proposed hybrid GCSVM technique as input and produces the output as a classification of two classes as positive samples and others as a negative sample.The SVM classification for both classes is represented in Eqs.(1) and (2) are:

    Is positive and,

    wheref(x) andf(y) are two classes of SVM and k is the outcome of these classes as a positive or negative.

    The above Tab.1 show the accuracy values of hybrid GCSVM segmentation method which shows the three datasets values including CSF fluid Leak, Low Grade Tumor, and CSF with low-grade tumor.In addition, the Training datasets of the hybrid GCSVM technique uses weight matrix and bias for each label of images to represent the hyper plane decision boundaries.In addition, hyper plane dimension is dependent on the number of features of data points in SVM classification.Hence,the proposed GCSVM technique is to develop the accuracy in images for the segmentation and preprocessing method etc., which improves not only the quality of images but reduces the redundancy which is the actual energy minimization in the segmentation process.Essentially, SVM calculates the hyper plane’s distance and generates the results in either a positive or negative class after detecting whether the CSF fluid and low-grade tumor are present or not in the datasets.In addition, features were extracted from the preprocessed image for further processing in this research.Each MRI image has its unique identification, and the system would match various features like shape, color, and intensity if the tumor or fluid in the images is present or not.All these features were extracted using SIFT.The SIFT works well, but still, there is some lack of identification to the nature of low-grade tumors or fluid.However, SIFT is able to show the sizes and features of the images but finding the range of fluid or un-shapeable tumors is hard to identify.This section prefers to apply the proposed HMkC approach to diagnose the process of tumor and CSF fluid.

    Table 1: Hybrid GCSVM model combination values of accuracy to maximize margin distance

    2.4 Hidden Markov Model (HMM) and K-mean Approach

    Tumor or fluid detection in medical MRI images is another problem of MRI images because identifying the CSF fluid (deposit the CSF fluid initially in the brain is unknown), and the low-grade tumor is not visible properly in the images.MRI is less effective in identifying the fluid or small size tumor in the brain until it stems from a sizable part.This problem is to be solved by the HMM, which is extended to the previous section.The HMM classifier is used to track the GCSVM technique to detect the CSF fluid and low-grade tumor that uses the spatial method of HMM.HMM combines probabilistic reasoning over time and space to the segment of the trained MRI datasets.In addition,the k-mean clustering algorithm also plays an important role in this section.The k-mean is used to store the classes of datasets in a cluster form in terms of classification.The proposed HMM approach creates the Markov chain and distributes the probability observed by the original state of the image.This section defines the transition state ofHMMasHMMis the best approach for identifying the fluid in the images.In this section, ICMM is applied to the length of the interval of sequence depending on the iteration condition to join the probability and calculate the probability range.After calculating the observed data, detection of CSF fluid in the image is obvious since the flow of CSF fluid leaks in one direction due to the nature of the liquid, and the tumor is located in different places of the brain.In this section, the area of fluid or tumor is identified by an iterated maximizing mode which finds the border of tumor or fluid in the construction of the image to detect it.In this research, the HMM classifier approach is combined with the GSVM technique in the k-NN algorithm to enhance the quality of the images.This approach develops the strategy of the k-NN algorithm to pick the nearest location of the“k”neighbor in terms of the k-mean cluster of the proposed technique.The hybrid GCSVM technique is to increase the quality of the images, and the feature vectors of the images consider the HMM of trained MRI images datasets statistically.This approach provides the best solution to check the availability and non-availability of tumor or Fluid by using the two transition states S = (S1, S2) such S1 show the presence of tumor or fluid and S2 show the absence of tumor or fluid.The method would be started in one of these two states and proceed smoothly from one state to another and generate an output of the probabilities of sequencing which belong to one category and the other category due to applying the condition of ICMM.If the Markov chainwere presently in state S1, then it would continue to the next state S2 with a specific probability.In order to procedure the Markov Transition matrix,this research firstly observe each of the given images by developing the ICMM to show the length of tumor (size) or range of fluid during the process of joining the probability.These probabilities identify the shape, size and range of fluid or low grade of tumor such ast, t+d, t+2d, t+3d,..., t+nd.Here, d is the length of sequence which shows the presence of tumor or fluid state.This transition state represents the probability of the state of the voxel at time strategy with the implementation of ICMM (maximum or minimum values sequentially depending upon the state condition) to the given state“t”.This HMM approach develops the main contribution with the utilization of the condition of the transition matrix such as the possible state of a voxel that has identified the Fluid or tumor or non-identified.

    3 Experimental Results and Discussion

    The experiments are conducted to evaluate the set research objectives.This chapter presents the simulation results and classification analysis of the proposed techniques of hybrid k-NN algorithm model.Different simulations are carried out to evaluate the performance of the proposed techniques to comparison of existing work and performance metrics.The efficiency of the proposed techniques is also compared with previous state-of-the-art k-NN algorithm.This section presents and discusses the results of the simulation formulated technique with statistical significance analysis.To evaluate the performance of the proposed hybrid GCSVM technique for classification of SVM and graph cut algorithm technique and image segmentation to diagnosing the tumor or CSF fluid in the image which is given below:

    Figs.2a-2c show the features of pixels corresponding to background, white matter, grey matter,tumor, and CSF Fluid in MRI images which is develop by the strategy of SVM and k-mean algorithm.This strategy is for classification of SVM using Gaussian radial basis kernel function which generates the values of healthy brain, CSF Fluid leak, Low-Grade tumors, CSF with Low-Grade tumors, and High-Grade tumors.Thus, simulation has been performed with various numbers of trained LFD datasets of MRI images from the range of 200 to 3000.Fig.2d shows the histogram of these simulation results of LFD datasets which indicate the range of CSF Fluid leakage in brain and Low-and Highgrade tumors with CSF.The consequence of this outcome is that the proposed SVM classification would support the next simulation technique to minimize the energy function of graph cut.The proposed SVM create the maximize the margin of classification that can SVM increase the efficiency of images while using the k-mean algorithm.

    Figure 2: (a)-(c) SVM classification for feature extraction of low-grade tumor with CSF fluid leak MRI datasets (d) histogram of SVM classification for feature extraction of low-grade tumor with CSF fluid leak MRI datasets

    The above Tab.2 shows the SVM classification simulation results values which maintain the quality of the images using the light field database images.Tab.2 show the brain tissue classification results of background, WM, GM, Tumor, and CSF in terms of Jaccard Similarity Coefficients for the simulated brain MRI images.

    Table 2: Datasets calculated values of SVM classification

    3.1 Implementation of Hybrid GCSVM Technique

    The simulation results of the max-flow/min-cut algorithms were used for the GrabCut segmentation method to detect both edges and regions.To perform segmentation, a graph is constructed, with nodes representing pixels in the images.The canny edge detection function (I, canny) was used in this study for edge detection in both the normalized cuts and the min cut/max flow algorithms.Firstly,the images get turned into 4D to 2D for noise removal and a median filter was applied to remove the impulses, as the images needed to smooth through a nonlinear 2D media filter with the help of‘MEDFILT1(I)’Including size 3×3.In addition, to segment the images, both nodes including sink and source nodes must be separated to each other and calculated the pixel’s values by determining the probability of the pixel node being part of the background or foreground region.Pixels in the same set as the source are labeled white, while pixels in the sink’s set are labeled black.For an input image of size 446×2 and 48761×2 for foreground and background, respectively, obtained by scaling the original image (see Figs.3a-3c) by a factor of 0.1 and removing the edges and convolving the images to obtain boundaries for connected and non-connected nodes.After developing the histogram, the number of connected nodes in the CSF fluid (CSF tissues) range was identified.Obtaining a binary image requires one threshold (T1) value that is required to segment between the white, grey matter and the grey matter of the CSF region.In min-max cut algorithm each region of the image was labeled to pixel intensity using the function of BWlabel.Furthermore, after calculating the sink and source nodes values need to be storing the values in“T and S”where T represents the sink node values and S represent the source node values as shown in the Tab.2.Furthermore, Fig.3 show the implementation results of all of three datasets of images with area of region, histogram of foreground and background values of all connected nodes, edge node detection values from sink and source node, Computation of Label Matrix (L) values result and, directed graph (g) node network connectivity, and also running time with complete details are mentioned as given below:

    Figure 3: (a)-(d) Calculated pixels’values foreground and background sink and source nodes

    The above area of region segmented values depends on the calculation of area of pixel targeted in the image for detection of the size of object in the image as this image indicate the value is high as per the targeted values of CSF fluid deposit in the brain as mentioned the detail in Tab.3.

    Table 3: Maximum flow results from nodes 1-10 of segmented area of region in image

    Object Area (pixels)1 44571

    3.2 Running Times of Trained Datasets

    Tab.4 shows the running times of the max-min cut flow algorithm, as shown in Figs.2-4 for various input sizes obtained by scaling down the input image.The scale factor refers to the image’s scale in one dimension, and rescale the image values is 0.1, so the actual scale down will be the square of the scale factor.This research considers the running times in milliseconds when tabulating the results of these experiments.The running time does not include the times for input and output.To better visualize the results, this research develops the table to plot the running times against the number of nodes in log scale as shown in Fig.5.

    Table 4: Running times for different input sizes in nodes and edges of the trained datasets

    Figure 4: Tumor identifies using the foreground and background sinks and source nodes

    Figure 5: The SIFT key-points for both the images are stored in features and identifying the object location of CSF fluid images datasets

    3.3 Implementation Results of Scale Invariant Feature Transform of Hybrid GCSVM Technique

    The simulated results of SIFT show the scale space of the initial images are repeated to select the trained datasets images by the use of the function of convolution of a variable-scale Gaussian, G (x, y,σ)with the input image (I).These results generate by the process of convolved with Gaussian’s function to produce the set of scale space images which shows on the left side.The adjacent Gaussian images are subtracted to produce the difference-of-Gaussian images on the right side and process is repeated after the Gaussian image is down sampled by a factor of 2.The below Figs.5 and 6 show their distance mapping and calculated values by comparing the pixel marked values of maximum and minimum difference of the Gaussian function images were detected to128 feature values and its neighbors in 3×3 at the recent region of the adjacent scale.

    Figure 6: Scale-invariant distance mapping point of object location of CSF fluid images datasets

    Datasets image Distance Value Low-grade tumor with CSF images Distance-D1 1.2118 Distance-D2 1.2168 Distance-D3 1.2753

    From the above Figs.5 and 6 show the SIFT is feature detection algorithm to detect and describe the local features of the images.The key-points or the points of interest are first selected.The 128-valued feature descriptor is used to match and recognize the object of interest in a new image based on the 128-valued feature descriptor already stored in the database.The Euclidean distance is used to determine the best matching points between the two images of their distance and mapping values of selected three database trained MRI images.

    3.4 Hidden Markov Model of k-mean Clustering Algorithm (HMk-NN) Technique

    The simulation results of the HMMkC used to detect the range of CSF fluid and low-grade tumor located place in MRI images.Fig.7 shows all of three datasets of images with the combination of hybrid GCSVM technique to merge with HMMkC technique to show the presence of tumor and CSF fluid that indicate the range of fluid and located the place of low-grade tumor.The classification of HMM for object detection is to consider each state of voxel of 4D MRI datasets to show the tumor or CSF fluid current situation with possible values of presence and absence.For instance, HMM is applied to tracking the segmentation over the time and space can be modeled with the help of transition matrix of both of two states.Fig.7 represent that the voxel was classified as a low-grade tumor size and CSF fluid leak range at the time (t-1) and the probability of the low-grade tumor and CSF fluid indicate that the timetis D1=0.96, D2=0.997, and D3=0.993 size.In addition, the HMM compute the posterior distribution of the transition state which is represented by filtering process.One shows the calculate the compute distance and other show the pixel intensity of both as an P(Xt/Xt +1).The pixel intensity state is applied for the classification of pixel intensity values at the time (t+1) and show the accuracy values of both tumor and CSF fluid leak in the brain are mentioned in Tab.5.Furthermore, ICMM are applied over HMM to retrieves the hidden information in the noisy images of trained datasets.The ICMM obtains datasets of images to estimate the ideal coefficients and therefore enables the selection of less visible images after linking the probabilities to each other with the transition matrix variable.Fig.10 show the simulation results of ICMM to show the clear vision of low-grade tumor and CSF occur in the MRI images after extracting the data in the images.Tab.6 show the compute accuracy values of ICMM over the HMM results of trained datasets which shows the improve results of the proposed technique.Below the given Figs.7 and 8 show the experimental results of ICMM over HMM.

    Figure 7: Combination of hybrid GCHMM Low-Grade tumor or CSF fluid detection in MRI images

    Table 5: Accuracy of HMM results over the trained datasets

    Figure 8: Results of ICMM over HMM classification of datasets

    3.5 Proposed Hybrid K-NN Algorithm

    The simulation results of the proposed hybrid k-NN algorithm classification of trained MRI datasets.The section of this research is used k-NN method for classification of hybrid GCHMkC technique for test sampling and training the samples of all datasets results.These results obtain the nearest neighbor location to conducting the k-NN classification of proposed hybrid technique.In k-NN algorithm, the values are assigning to the test samples with labels of selected nearest neighbor for each test sample by the setting of the optimal value of k in the training datasets.It is applied for fixed the value of k for all test sample of proposed hybrid technique that have been focused the different value of k with different samples results generate.Below the Tab.6 show the accuracy, sensitivity and specificity, and computational results mentioned are given below:

    Table 6: k-NN algorithm classification results of accuracy, sensitivity, and specificity, and computational time

    The given below Fig.9 show the results of all of three database images results which show the k-NN classification with better results.

    Figure 9: k-NN algorithm classification results of trained datasets images

    The above Fig.10 shows the combination of hybrid GCHMkC technique including four different methods such as SVM classification; Grabcut segmentation, Hidden Markov Model, and k-mean clustering algorithm that is combine together and convert into hybrid model.This hybrid model creates the novel results of hybrid k-NN algorithm classification of low-grade tumor and CSF fluid leakMRI images of accuracy of 91.1%, sensitivity of 99.9%, specificity of 99.9%, and computational time is 14.99 s.These results generate the better accuracy compare than the previous results.These parameters detect tumor, tumor size, identify the location CSF inside the brain.This hybrid model helps them both common man and doctors as well to get easily identify the low-grade tumor and CSF inside the brain and diagnose the tumor in the early stage that will try to save the human life.

    Figure 10: Hybrid k-NN algorithm of proposed technique of GCHMkC

    4 Pseudo Code of Hybrid GCHMkC Technique in k-NN Algorithm

    The pseudo code of hybrid GCHMkC technique in k-NN algorithm is based on four different methods that combine together in this algorithm to develop this GCHMkC technique is given below:

    5 Conclusion

    The technique was implemented on the MRI images of CSF fluid with low-grade tumor to improve the performance on the trained datasets by the proposed technique.The experimental simulations for GCSVM implemented on GrabCut segmentation in trained MRI images and using scale invariant feature transform to extract the data that can HMM refine and reconstruct the images.However, the experimental simulations for HMkC are implemented with ICMM to generate the sequence in the images and join the HMM probability that can identify the range of low grade tumor or CSF fluid liquid in HMM whereas the k-mean store the data of trained MRI images.Furthermore,k-NN algorithm for classification of hybrid GCHMkC technique for test sampling and training the samples of all datasets results which gives the better results which help to pick the nearest position of tumor and CSF fluid easily after implementing this novel technique.The motive of this research is to develop the unique model to identify the low-grade tumor and CSF fluid leak in the initial phase of starting the brain cancer.The sensitivity and specificity is conducted by the method of prevalence of diseases and computational time is the execution time of software which generates these results.

    Funding Statement:Taif University Researchers Supporting Project Number (TURSP-2020/98), Taif University, Taif, Saudi Arabia.

    Conflicts of Interest:The authors declare that they have no conflicts of interest to report regarding the present study.

    亚洲人与动物交配视频| 国产午夜精品论理片| 九九在线视频观看精品| 美女免费视频网站| 日韩强制内射视频| 欧美色视频一区免费| 一卡2卡三卡四卡精品乱码亚洲| 赤兔流量卡办理| 三级男女做爰猛烈吃奶摸视频| 欧美又色又爽又黄视频| 一级a爱片免费观看的视频| 国产一区二区亚洲精品在线观看| 好男人在线观看高清免费视频| 直男gayav资源| 看免费成人av毛片| 精品人妻视频免费看| 成年女人永久免费观看视频| 欧美在线一区亚洲| 亚洲人成网站在线播| 国产伦一二天堂av在线观看| av中文乱码字幕在线| 亚洲av美国av| 嫩草影院入口| 又爽又黄a免费视频| 亚洲欧美日韩东京热| 国产三级中文精品| a级毛片a级免费在线| 1024手机看黄色片| 美女被艹到高潮喷水动态| 国产精品亚洲一级av第二区| 国产私拍福利视频在线观看| 午夜免费激情av| 免费黄网站久久成人精品| 国产伦在线观看视频一区| 大型黄色视频在线免费观看| 免费高清视频大片| 久久九九热精品免费| 直男gayav资源| 色综合色国产| 久久精品国产自在天天线| 亚洲美女黄片视频| 国产精华一区二区三区| 嫩草影院入口| 国产日本99.免费观看| av黄色大香蕉| 午夜免费男女啪啪视频观看 | av在线老鸭窝| 91在线观看av| 真人做人爱边吃奶动态| 亚洲国产精品合色在线| 午夜日韩欧美国产| 小说图片视频综合网站| 婷婷精品国产亚洲av在线| 最新中文字幕久久久久| 精品久久国产蜜桃| 亚洲aⅴ乱码一区二区在线播放| 女人十人毛片免费观看3o分钟| 深爱激情五月婷婷| 色哟哟·www| 中文字幕熟女人妻在线| 九九热线精品视视频播放| 亚洲国产欧美人成| 日本免费a在线| 男女边吃奶边做爰视频| 国产一区二区激情短视频| 听说在线观看完整版免费高清| 制服丝袜大香蕉在线| 色播亚洲综合网| 日本一二三区视频观看| 麻豆国产av国片精品| 美女 人体艺术 gogo| 老司机福利观看| 午夜精品一区二区三区免费看| 黄色配什么色好看| 国产精品一及| 人妻少妇偷人精品九色| a级一级毛片免费在线观看| 国内少妇人妻偷人精品xxx网站| 波多野结衣高清作品| 给我免费播放毛片高清在线观看| 波多野结衣巨乳人妻| 大型黄色视频在线免费观看| 观看美女的网站| 国产精品伦人一区二区| 亚洲国产精品合色在线| 亚洲美女视频黄频| 国产精华一区二区三区| 日韩欧美在线乱码| 亚洲精品成人久久久久久| 亚洲av熟女| 亚洲av不卡在线观看| 欧美激情久久久久久爽电影| 午夜a级毛片| 99国产精品一区二区蜜桃av| 级片在线观看| 国产精品一区二区性色av| 国产亚洲91精品色在线| 国产视频内射| 国产免费男女视频| 精品久久久久久久久亚洲 | h日本视频在线播放| 99久久精品热视频| 简卡轻食公司| 欧美绝顶高潮抽搐喷水| 国产蜜桃级精品一区二区三区| 欧美xxxx黑人xx丫x性爽| 日韩欧美国产在线观看| 国产精品综合久久久久久久免费| 九九热线精品视视频播放| 日本撒尿小便嘘嘘汇集6| 久久久久免费精品人妻一区二区| 3wmmmm亚洲av在线观看| 欧美在线一区亚洲| 国产综合懂色| 久久热精品热| 最近中文字幕高清免费大全6 | 亚洲专区中文字幕在线| 俄罗斯特黄特色一大片| 国产久久久一区二区三区| 免费av不卡在线播放| 久久精品国产清高在天天线| 成人av在线播放网站| av专区在线播放| 日韩一本色道免费dvd| 听说在线观看完整版免费高清| 久久久国产成人精品二区| 黄色配什么色好看| 99久久精品热视频| 午夜精品久久久久久毛片777| 小蜜桃在线观看免费完整版高清| eeuss影院久久| 男女视频在线观看网站免费| 精品久久久久久久久亚洲 | 国产精品亚洲一级av第二区| a级毛片免费高清观看在线播放| x7x7x7水蜜桃| 日韩国内少妇激情av| 国产高清激情床上av| videossex国产| 国产高清三级在线| av中文乱码字幕在线| 亚洲七黄色美女视频| 久久精品国产亚洲av天美| 九九在线视频观看精品| 亚洲狠狠婷婷综合久久图片| 高清在线国产一区| 又粗又爽又猛毛片免费看| 日韩欧美免费精品| 亚洲人成网站在线播| 成人特级黄色片久久久久久久| 国产亚洲91精品色在线| 婷婷丁香在线五月| 亚洲成人久久性| 中文字幕熟女人妻在线| 国产aⅴ精品一区二区三区波| 国产高潮美女av| 狂野欧美激情性xxxx在线观看| 干丝袜人妻中文字幕| 99久久九九国产精品国产免费| 久久99热这里只有精品18| 精品久久久久久久末码| 中文字幕av在线有码专区| 日韩欧美精品v在线| 欧美黑人巨大hd| 日本一本二区三区精品| 国产三级在线视频| 色精品久久人妻99蜜桃| 人人妻人人看人人澡| 97人妻精品一区二区三区麻豆| 亚洲精品影视一区二区三区av| 久久久久久久亚洲中文字幕| av在线观看视频网站免费| 99久久无色码亚洲精品果冻| 成人综合一区亚洲| 美女免费视频网站| 午夜爱爱视频在线播放| 成年版毛片免费区| 国产高清视频在线观看网站| 成熟少妇高潮喷水视频| 成人特级黄色片久久久久久久| 国产黄a三级三级三级人| 欧美3d第一页| 国产一区二区三区视频了| 亚洲狠狠婷婷综合久久图片| 真实男女啪啪啪动态图| 日本色播在线视频| 午夜精品久久久久久毛片777| 亚洲性久久影院| 国产免费av片在线观看野外av| 在线免费十八禁| av在线亚洲专区| 久久精品国产亚洲av香蕉五月| 久久久国产成人免费| 亚洲性久久影院| 91麻豆精品激情在线观看国产| 乱系列少妇在线播放| 久久精品国产亚洲av涩爱 | 欧美又色又爽又黄视频| 禁无遮挡网站| av专区在线播放| 91午夜精品亚洲一区二区三区 | 99久国产av精品| 亚洲av美国av| 毛片一级片免费看久久久久 | 久99久视频精品免费| 日韩一区二区视频免费看| 国语自产精品视频在线第100页| 国产精品亚洲美女久久久| 成人欧美大片| 神马国产精品三级电影在线观看| 国产精品自产拍在线观看55亚洲| 日本撒尿小便嘘嘘汇集6| 国产精品乱码一区二三区的特点| 一级a爱片免费观看的视频| 69av精品久久久久久| 91精品国产九色| 伦精品一区二区三区| 国产男靠女视频免费网站| 亚洲av日韩精品久久久久久密| 麻豆成人午夜福利视频| 免费人成在线观看视频色| 夜夜看夜夜爽夜夜摸| 国产高清激情床上av| 亚洲精品影视一区二区三区av| 欧洲精品卡2卡3卡4卡5卡区| 国产乱人伦免费视频| 国产精品av视频在线免费观看| 99国产精品一区二区蜜桃av| 看片在线看免费视频| 成人毛片a级毛片在线播放| 十八禁网站免费在线| 亚洲欧美日韩高清专用| 啦啦啦韩国在线观看视频| 91久久精品电影网| 成年女人永久免费观看视频| 51国产日韩欧美| 真人做人爱边吃奶动态| 麻豆久久精品国产亚洲av| 久久人人爽人人爽人人片va| 亚洲成人久久爱视频| 小蜜桃在线观看免费完整版高清| 久久国产乱子免费精品| 午夜福利在线观看吧| 啦啦啦韩国在线观看视频| 999久久久精品免费观看国产| 丰满乱子伦码专区| 91麻豆av在线| 欧美日韩瑟瑟在线播放| 97人妻精品一区二区三区麻豆| 免费观看人在逋| 免费看日本二区| 成人综合一区亚洲| 国产高潮美女av| 99久久中文字幕三级久久日本| 嫁个100分男人电影在线观看| 高清在线国产一区| 不卡视频在线观看欧美| 久久精品人妻少妇| 少妇的逼好多水| 欧美性感艳星| 欧美3d第一页| 看片在线看免费视频| 日韩一本色道免费dvd| 成人特级av手机在线观看| 久久久久久久久中文| 婷婷色综合大香蕉| 国内精品久久久久久久电影| ponron亚洲| 久久6这里有精品| 久久久久久久久中文| 国产精品,欧美在线| 国产单亲对白刺激| 国产av在哪里看| 最后的刺客免费高清国语| 91狼人影院| 国产亚洲91精品色在线| 国产色爽女视频免费观看| www日本黄色视频网| 可以在线观看的亚洲视频| 又紧又爽又黄一区二区| 夜夜爽天天搞| 看免费成人av毛片| 日本免费a在线| 免费一级毛片在线播放高清视频| 久久人人爽人人爽人人片va| 又黄又爽又刺激的免费视频.| 男人舔女人下体高潮全视频| 国产v大片淫在线免费观看| 久久久久久九九精品二区国产| 97热精品久久久久久| 亚洲成av人片在线播放无| 亚洲18禁久久av| 亚洲不卡免费看| 精品久久久久久,| 亚洲最大成人手机在线| 亚洲中文字幕日韩| 亚洲精华国产精华精| 久久精品国产亚洲av涩爱 | 神马国产精品三级电影在线观看| 午夜福利在线观看吧| 欧美最黄视频在线播放免费| 欧美又色又爽又黄视频| 日韩欧美免费精品| 老司机福利观看| 久久精品91蜜桃| 精品午夜福利在线看| 在线免费十八禁| 搡老熟女国产l中国老女人| 久久精品影院6| 午夜久久久久精精品| 18禁黄网站禁片免费观看直播| 国产成人av教育| 国产午夜福利久久久久久| 精品久久久久久久末码| 国产毛片a区久久久久| av在线天堂中文字幕| 99精品在免费线老司机午夜| 国产91精品成人一区二区三区| 成熟少妇高潮喷水视频| 国产不卡一卡二| 色视频www国产| 欧美+亚洲+日韩+国产| 1024手机看黄色片| 国产女主播在线喷水免费视频网站 | 日本 av在线| 久久草成人影院| av在线观看视频网站免费| 亚洲成人久久爱视频| 欧美区成人在线视频| 观看美女的网站| 精品人妻一区二区三区麻豆 | 黄色一级大片看看| 有码 亚洲区| 成人二区视频| 不卡一级毛片| av视频在线观看入口| 国产又黄又爽又无遮挡在线| 午夜免费成人在线视频| 一边摸一边抽搐一进一小说| 欧美高清成人免费视频www| 午夜福利欧美成人| 精品人妻熟女av久视频| 国产乱人伦免费视频| 变态另类丝袜制服| 欧美另类亚洲清纯唯美| 午夜福利高清视频| 性插视频无遮挡在线免费观看| 啦啦啦观看免费观看视频高清| 99国产极品粉嫩在线观看| 69av精品久久久久久| 99热这里只有是精品50| 一级a爱片免费观看的视频| 亚洲va在线va天堂va国产| 免费一级毛片在线播放高清视频| 日韩精品有码人妻一区| 无人区码免费观看不卡| 欧美日韩综合久久久久久 | 夜夜夜夜夜久久久久| 国语自产精品视频在线第100页| 十八禁网站免费在线| 免费看av在线观看网站| 国产一级毛片七仙女欲春2| 舔av片在线| 欧美高清性xxxxhd video| 1000部很黄的大片| 久久亚洲真实| 天天一区二区日本电影三级| 两个人视频免费观看高清| 欧美人与善性xxx| 亚洲狠狠婷婷综合久久图片| 尾随美女入室| 免费观看人在逋| 国产伦在线观看视频一区| 听说在线观看完整版免费高清| 18+在线观看网站| a级毛片a级免费在线| 少妇的逼水好多| 国产精品三级大全| 久久久久久久久中文| 成人特级av手机在线观看| 看片在线看免费视频| 国产一区二区激情短视频| 日韩一区二区视频免费看| 有码 亚洲区| 国产伦精品一区二区三区四那| 欧美精品国产亚洲| 国产精品电影一区二区三区| 女生性感内裤真人,穿戴方法视频| 日日夜夜操网爽| 国产黄a三级三级三级人| 欧美最黄视频在线播放免费| 亚洲av日韩精品久久久久久密| 99久久精品一区二区三区| 亚洲熟妇熟女久久| 国产激情偷乱视频一区二区| 亚洲无线在线观看| 国产成人一区二区在线| 日日干狠狠操夜夜爽| 亚洲最大成人中文| 免费搜索国产男女视频| 小说图片视频综合网站| 在现免费观看毛片| ponron亚洲| 三级毛片av免费| av天堂在线播放| 麻豆av噜噜一区二区三区| 国产精品无大码| 欧美中文日本在线观看视频| 久久精品国产自在天天线| 欧美另类亚洲清纯唯美| 高清日韩中文字幕在线| 淫妇啪啪啪对白视频| 天堂√8在线中文| 99精品在免费线老司机午夜| 亚洲狠狠婷婷综合久久图片| 不卡视频在线观看欧美| 色精品久久人妻99蜜桃| 欧美日韩黄片免| 最近在线观看免费完整版| 成人毛片a级毛片在线播放| 51国产日韩欧美| 亚洲真实伦在线观看| 性欧美人与动物交配| 一级a爱片免费观看的视频| 欧美成人免费av一区二区三区| avwww免费| 在线播放国产精品三级| 亚洲熟妇中文字幕五十中出| 看黄色毛片网站| 久久久精品欧美日韩精品| netflix在线观看网站| 日本免费a在线| 国产精品一区二区三区四区免费观看 | 免费搜索国产男女视频| 在线天堂最新版资源| 男人和女人高潮做爰伦理| 我的老师免费观看完整版| 国产真实乱freesex| 亚洲精品亚洲一区二区| 亚洲精品456在线播放app | 啦啦啦观看免费观看视频高清| 国产一区二区在线av高清观看| 看黄色毛片网站| 在线观看美女被高潮喷水网站| 看免费成人av毛片| 久久久久九九精品影院| 久久热精品热| 老司机午夜福利在线观看视频| 波多野结衣高清无吗| 非洲黑人性xxxx精品又粗又长| av中文乱码字幕在线| 中出人妻视频一区二区| 一进一出好大好爽视频| 国产精品电影一区二区三区| 国内毛片毛片毛片毛片毛片| 99久久九九国产精品国产免费| www.www免费av| 国产精品一区二区性色av| 亚洲av免费高清在线观看| 亚洲专区中文字幕在线| 成年女人毛片免费观看观看9| 久久亚洲精品不卡| 一本精品99久久精品77| АⅤ资源中文在线天堂| 在线观看舔阴道视频| 男女之事视频高清在线观看| 黄色日韩在线| 99在线视频只有这里精品首页| 亚洲av中文av极速乱 | 香蕉av资源在线| 啦啦啦韩国在线观看视频| 乱人视频在线观看| 国语自产精品视频在线第100页| 午夜福利视频1000在线观看| 免费av不卡在线播放| 免费搜索国产男女视频| 日韩精品中文字幕看吧| 一个人免费在线观看电影| 99精品在免费线老司机午夜| 一进一出好大好爽视频| 性欧美人与动物交配| 成人国产一区最新在线观看| 亚洲最大成人手机在线| 女同久久另类99精品国产91| 久久久久久久精品吃奶| 一个人看视频在线观看www免费| 亚洲最大成人av| 国产探花在线观看一区二区| 欧美黑人欧美精品刺激| 国产在视频线在精品| 精华霜和精华液先用哪个| 日本撒尿小便嘘嘘汇集6| 日韩一本色道免费dvd| 精品一区二区三区视频在线| 免费观看在线日韩| 性色avwww在线观看| 三级男女做爰猛烈吃奶摸视频| 啪啪无遮挡十八禁网站| 看免费成人av毛片| 日日干狠狠操夜夜爽| 伊人久久精品亚洲午夜| 深爱激情五月婷婷| 欧美日本视频| 亚洲av熟女| 永久网站在线| 成人鲁丝片一二三区免费| 国产主播在线观看一区二区| 99精品在免费线老司机午夜| 日本a在线网址| 全区人妻精品视频| 在线看三级毛片| 日韩精品有码人妻一区| 啦啦啦韩国在线观看视频| 观看美女的网站| 熟女电影av网| xxxwww97欧美| 久久精品国产清高在天天线| 免费av观看视频| 99久久无色码亚洲精品果冻| 成人毛片a级毛片在线播放| 蜜桃亚洲精品一区二区三区| 欧美黑人欧美精品刺激| 亚洲,欧美,日韩| videossex国产| 又粗又爽又猛毛片免费看| 亚洲avbb在线观看| 麻豆国产97在线/欧美| 少妇熟女aⅴ在线视频| 国内精品宾馆在线| 少妇人妻一区二区三区视频| 国产精品一区二区免费欧美| 黄色视频,在线免费观看| 在线a可以看的网站| 国产精品美女特级片免费视频播放器| eeuss影院久久| 大又大粗又爽又黄少妇毛片口| 校园人妻丝袜中文字幕| 日本黄色视频三级网站网址| 亚洲成a人片在线一区二区| 国产精品久久久久久精品电影| 赤兔流量卡办理| 天天躁日日操中文字幕| 久久精品人妻少妇| 日本三级黄在线观看| 精品久久久久久久久av| 99久久九九国产精品国产免费| 在线免费十八禁| 国产不卡一卡二| 窝窝影院91人妻| 两个人视频免费观看高清| 悠悠久久av| 欧美日韩亚洲国产一区二区在线观看| АⅤ资源中文在线天堂| 成人国产综合亚洲| 成人av在线播放网站| 少妇丰满av| 国产精品电影一区二区三区| 欧美在线一区亚洲| 中文字幕av成人在线电影| 亚洲不卡免费看| 亚洲无线在线观看| 女同久久另类99精品国产91| 色视频www国产| 国产成年人精品一区二区| 不卡视频在线观看欧美| 在线免费十八禁| 国内揄拍国产精品人妻在线| 成人精品一区二区免费| 成人无遮挡网站| 2021天堂中文幕一二区在线观| 精品人妻一区二区三区麻豆 | 一区二区三区四区激情视频 | 变态另类成人亚洲欧美熟女| 久久久久性生活片| 99热只有精品国产| 日本熟妇午夜| 国产美女午夜福利| 他把我摸到了高潮在线观看| 国产伦在线观看视频一区| 亚洲av免费高清在线观看| 五月伊人婷婷丁香| 看片在线看免费视频| 一区二区三区免费毛片| 久久人人精品亚洲av| 亚洲人成网站高清观看| 亚洲av第一区精品v没综合| 一进一出抽搐gif免费好疼| 亚洲av电影不卡..在线观看| 亚洲av日韩精品久久久久久密| av福利片在线观看| 日本黄色片子视频| 免费高清视频大片| h日本视频在线播放| 欧美色欧美亚洲另类二区| 88av欧美| 少妇高潮的动态图| 一级黄片播放器| 99热只有精品国产| 欧美激情久久久久久爽电影| 午夜激情欧美在线| 在线免费观看的www视频| 欧美日韩瑟瑟在线播放| 三级毛片av免费| 亚洲一级一片aⅴ在线观看| 欧美激情国产日韩精品一区| 简卡轻食公司| 色综合亚洲欧美另类图片| 伦精品一区二区三区| 国产真实乱freesex| 日韩欧美一区二区三区在线观看| 国产伦在线观看视频一区| 麻豆成人午夜福利视频| 黄色女人牲交| 伦精品一区二区三区| 午夜爱爱视频在线播放|