A performance study of image segmentation techniques

A Performance Study of Image Segmentation Techniques Arti Taneja1 Research Scholar,Amity Institute of Information Technology 1 [email protected] Dr. Priya Ranjan2 Professor, Amity university Uttar Pradesh,Noida 2 [email protected] Abstract— Image based applications such as target tracking, tumor detection, texture extraction requires an efficient image segmentation process. The partitioning of image into various non- overlapping distinct regions refers the image segmentation. Various segmentation techniques like edge, threshold, region, clustering and neural network are involved in the effective image analysis. The efficiency of the segmentation process improved with the help of several algorithms, namely, active contour, level set, Fuzzy clustering and K-means clustering. This paper analyses the performance of algorithms for image segmentation in detail. Intensity and texture based image segmentation is the two levels of the level set method. The combination of both intensity and texture based image segmentation provides better results than the traditional methods. The detailed survey of segmentation techniques provides the requirement of the suitable enhancement method that supports both intensity and texture based segmentation for better results. The comparison between the traditional image segmentation techniques are illustrated. Index Terms— Active Contour Model, Fuzzy-C-Means (FCM), Image Segmentation, Gaussian Mixture Model (GMM), K-Means Clustering, Level Set methods. I. INTRODUCTION Computer vision applications require an image segmentation to extract the meaningful regions of the image. The objective of efficient image segmentation is independent partitioning of regions that are visually different, meaningful to image characteristics and properties. The classification of image segmentation techniques [1] based on the detection performance are edge, threshold, region, fuzzy and neural network based. The edge detection [2] based recognition of real images reduces the false hit ratio. Threshold based techniques [3] such as mean, p-tile, edge maximization and visual applied to improve the performance of image segmentation. The region based [4], fuzzy based Gaussian Mixture Model (GMM) [5] and neural network [6] applied to segmentation in order to analyze the curvature regularity, energy function and noise effects in the image. The detailed description of image segmentation techniques based on the various factors discussed in this paper. The refinement of boundary is an important requirement in image segmentation. Contour based approaches [7-9] applied on various images in order to handle the topology changes, noise, streaks and faint spots. The active contour segmentation extends into morphological nature for accurate refinement of boundary with more energy consumption. The split Bergman method efficiently minimizes the energy function. Contour based approaches ignores the spatial relationship between the colors of images, thereby segmented 978-1-4673-7231-2/15/$31.00 ©2015 IEEE Dr.Amit Ujjlayan3 Professor,GautamBudha University,Greater Noida,India 3 [email protected] regions are fragmented. Fuzzy based clustering approaches are used for improvement in compactness that effectively reduces the limitation of fragmentation. Histogram Thresholding Fuzzy-C-Means (HTFCM) [10], adaptive FCM [11] and improved FCM [12] algorithms are used to improve the compactness of clusters. The evolution of clustering algorithms provides the parameters such as the number of clusters and the location of cluster centroid, which plays the major role in image segmentation. Hence, image segmentation involves K-means clustering algorithms [13-15] to predict the number of clusters and location of cluster centroid. The accuracy and reproducibility are enhanced with minimum execution time by using K-means approaches. Level set methods [16-21] involves the combination of discriminative classification models and distance transforms to predict the snap closest to true object with less local minima. The preliminary stage at diagnosis and brain image segmentation depends upon various factors such as homogeneity, contrast, noise and inequality content. The GMM reviews the factors on brain MRI image segmentation. The multi-scale graph cut approach and Maximum a Posteriori (MAP) principles are applied to image segmentation with pixel density functions. Multimodal image registration processing speed improved by multi-atlas based methods [2227]. The textural analysis enhanced the object recognition system with sensing and collaborative information. Texture descriptors analyze the filter responses for effective discrimination. The texture based image segmentation utilizes the Convolutional Neural network (CNN) to recover the 3 D scene layout of the road scene image. The deformable models used both intensity and edge based feature traditionally. The inclusion of texture information increases the effectiveness of proposed image segmentation [28-35]. This paper presents the detailed analysis of the techniques involved in image segmentation and the performance parameters predicted in the traditional research works. The paper is organized as follows. Section 2 describes the overview of major categories of an image segmentation. Section 3 presents various image segmentation techniques. Section 4 describes results and discussion about the traditional segmentation techniques. Section 5 provides proposed work and section 6 discusses conclusions. II. IMAGE SEGMENTATION TECHNIQUES The detailed analysis of image segmentation categories [1-5] presented in the literature works to identify edge, region depends upon pixel and grouping of pixels with improved processing speed. The image segmentation techniques and associated equations are described as follows: A. Active Contour model Image segmentation utilizes the contour models to describe the boundary regions and improve the accuracy. The boundary refinement is an important process in contour model. Three techniques used for boundary refinement analysis as follows: 1. Constrained active contour The convex active contour model with an edge detection function [7] on image domain described by min 0 ≤ u ≤1 (∫ g b ∇ u dx + λ ∫ h r udx ) (1) and Euclidean Initially, probabilistic function distance are used for contour initialization. The formulation of regional term with the probabilistic method of foreground ( ) and background ( ) x ( x ) ( x ) ( ) = − (2) hr PF PB Finally, the boundary term described by equation (1) as follows: weighed by the function u and the weight 1 (3) g b (x) = 2 1 + ∇ I (x ) Here, I(x) describe the intensity of image pixels. The incorporation of probability function into the edge detection function improve the performance. The new weight function comprises trade off factor ( β ) and weight function of edge ) described as, and contour ( (4) g b = β . g c + (1 − β ). g e The equation (4) returns the values from 0 to 1 for most likely edges. The image analysis based on edge detection model in (4) effectively performs without consideration of noise effects. 2. Convex energy function The weight total variation of constrained active contour incorporates the energy model to detect the boundaries easily on the presence of small amount of Gaussian noise. The Local Gaussian Distributing Fitting (LGDF) analyzes the Gaussian noise effectively. The LGDF is pre based new energy function with the difference of gradient flow coefficients described as (5) E (ϕ ) = ∫ ∇ (ϕ ) dx + ∫ ϕ (x ) r (x ) dx The segmentation model based on equation (5) extracts the boundaries of an image with Gaussian noise. 3. Unsupervised segmentation The Chan-Vese (CV) model [8] for local intensity analysis describes the region of an image expressed by the partial differential equation as follows: ⎡ ⎛ ∇ϕ ⎞ 2 ∂ϕ ⎟ − + − + + λ 1− u − c − = δ (ϕ ) ⎢ µ div ⎜ 1 1 ⎜ ∇ ϕ ⎟ λ 1 u1 c1 ∂t ⎢⎣ ⎝ ⎠ ( ) ( 2⎤ ) ⎥⎥ (6) ⎦ Here, describes the average intensities of image described by the level set function with Dirac delta function. The abnormalities in group of image regions severely affects the quality of segmentation. B. Fuzzy Clustering The Fuzzy C-means clustering methods enhances the compactness of regions to separate the similar and dissimilar regions of an image. The FCM [10] based image analysis classified into three categories as follows: 1. Histogram thresholding The objective function contains membership degree and distance between pixels in cluster for FCM expressed as N W ij M = ∑ ∑ u ijm d 2 ij i =1 j =1 (7) The intensity inhomogeneities introduced in micro imaging system affect the segmentation quality. Hence, the improvement in FCM needed to improve the quality. 2. Adaptive FCM The objective function [11] comprises observed image and weighing function intensities , cluster centers required to improve the quality of an image expressed as follows: NC 2 ( q J IAFCM = ∑ ∑ uik yi − g i ck + λ ∑ g i −( H ∗g )i i∈D k =1 i∈D ) 2 (8) The membership function and cluster center are expressed as follows: u ik NC ∑ l=1 ck = k y i− g ic ∑ u ikq G i∈ D ∑ u i∈ D q ik i G 2 q −1 − y i− g ic = − l (9) 2 q −1 yi (10) 2 i The relationship between gain field and convolution kernel defined by (11) g i = ( H * g )i , i ∈ D The gain field in terms of membership function represented as NC g i = ∑ u q ik yi,ck ∑ u q ik ck ,ck k =1 NC k =1 (12) The image segmentation based on AFCM improves the classification accuracy by detecting the intensity inhomogeneities. The robustness to noise and outliers is poor by using AFCM algorithm. 3. Kernel metric The inclusion of kernel metric in the objective function improves the robustness of image segmentation. The kernel measure (K) [12] depends upon the parameters of kernel bandwidth ( ) and dimension of image vectors (d) ⎛ d a⎞ t = − ⎜ ∑ xi − y i ⎟ ⎜ ⎟ ⎠ ⎝ i =1 ⎛t ⎞ K ( x, y ) = exp⎜ ⎟ ⎝σ ⎠ b (13) (14) The cluster center function in (10) rewritten as by using kernel metric as N ( ∑ u mki K ( x i , v k ) x i ck = i =1 N ( ∑ u mki K ( x i , v k ) i =1 ) ) (15) The reconstruction of the original image at the point of convergence produces the accurate result. C. K-means Clustering The unsupervised learning algorithm for accurate extraction of tumor from an image is K-means clustering. In K-means process [14] , the distribution of vertices on K-means clusters on the basis of Poisson process refers to Prim’s trajectory [13]. Prim’s algorithm relates the k to Probability of False Alarm (PFA) in L-dimensional Euclidean space defined by L⎞ ⎛ PFA (k , ∈ ) = ⎜ 1− e − λ ∈ ⎟ ⎠ ⎝ k (16) The matrix of labels are estimated by constructing minimum spanning tree using Prim’s algorithm. An Automated 2-Dimensional K means algorithm (A2DKM) [15] eliminates the need of number of clusters. D. Level Set Methods Level set methods consider the topological changes to describe the curves. The deformation of the curve represented by Partial differential equation [16] as follows: ∂ ϕ (x , y , t ) = F ∇ϕ ∂t (17) The evolving curve in image segmentation stopped at an object boundary by following two ways, namely, application of edge-stopping function and minimization of the energy function. Due to the presence of noise or insignificant image gradients in evolving curve, shrinking or growing based on the sign of F occur termed as one way curve evolution problem. To resolve this problem, the analysis of the curve based on an energy criterion introduced. The energy function depends , area , upon the various parameters such as length inside and outside region of curve represented as E = µ ⋅ C + v ⋅ C in + F (C in ) + F (C out ) (18) The segmentation process involves the more than one two regions to be segmented. Hence, multilevel set multi-phase model [18] included in the level set energy function as N E (ϕ , c , b ) = ∫ ∑ e i ( x ) M i (ϕ (x ))dx (19) i =1 2 e i = ∫ K ( y − x ) I (x )− b ( y )c i dy (20) Here K represents the kernel function defined by (14). The carried out to energy minimization with respect to predict the optimal cluster center and boundary. The level set function in (17) redefined by using the probabilistic parameters [19] to represent the region boundary and curvature as ∂ ϕ (x , y , t ) (21) = − [α ⋅ R ( x , t ) + β ⋅ B (x , t ) + γ ⋅ C (x , t )] ∇ ϕ ∂t Gaussian filter effectively smoothens the initial pixel wise probabilistic values. Local classifiers and boundary refinement methods are used for effective refinement process. Level set formulation extends the segmentation application to localization of optical disk in retinal images [21], tracking of non-linear shapes [20] and liver tumor [17]segmentation. Level set models excludes the information other the boundary and the intensity inhomogeneity. E. Intensity Based Segmentation To analyze the inhomogeneity, the image can be modelled as normally distributed noise with zero mean in intensity based image segmentation [27] as follows: I ( x ) = b (x ) I o ( x ) + n (x ) (22) The inclusion of density functions in the equation (22) to analyze the inhomogeneity provides the following model ⎛ p ⎜ I ( x ) | {χ , b , c , σ } α ⎜ ⎝ ( ∏ p I (x ) | {χ , b , c ,σ }π ρ y∈ w x x ( y) )⎞⎟⎟ (23) ⎠ The adaptive weights depend upon the distance from y to x with the truncated kernel function as (24) π x ( y) = K (y − x) Based on the density function and the adaptive weights, the intensity based image feature such as DiscontinuityHomogeneity (DH) ratio [24] of discontinuity descriptor to intensity defined by DH f s ⎛⎜ x , r ⎜ ⎞ ⎟ ⎟ ⎠ ⎝ = var s (I ) ⎛ ⎞ ⎜ x r ⎟ ⎜ ⎟ ⎝ ⎠ , (25) (I ) The segmentation results further improved by adding Image Derived Attributes (IDA) [26] in the intensity models. The performance measures to describe the IDA are Dice Similarity Coefficient (DSC) and Mean Absolute Distance (MAD). The maximal similarity from the Normalized Cross Correlation achieved in atlas based models [23] the ratio of NCC of the similarity measure of atlas defined by ri = NCC (T , A i D M i ) max i NCC (T , A i D M i ) (26) The review of Gaussian Mixture Models (GMM) [22] based on similarity index presented on brain MRI images. The intensity based models laid a stone to reduce the coding length and accurate recognition of an object. F. Texture Based Segmentation The optimal segmentation based on textures efficiently reduces the coding length [30]. The total coding length of an image depends upon the length of particular region and the whole boundary of an image defined by k S L w , e (R ) = ∑ L w , e (R i ) + i =1 1 B (R i ) 2 (27) The reduction of length defined by (23) performed by using agglomerative approximation. The uniformity nature of regions extends the segmentation approach into recognition of with road sign image [28]. The histogram of uniformity each pixel and the corresponding intensity level defined by, L (28) U = ∑ p 2 (j) j=1 N y (i) = ∑ w j =1 j x j (i ) (29) An optimal solution corresponding to the estimating of minimum variance of weight for accurate road scene recognition is given by, ( W = Σ −1 I I T Σ −1I ) −1 (30) Finally, higher value of likelihood road side areas is computed. The landmarks in target image represented by another texture based segmentation termed as game theoretic approach [29]. The intensity and shape likelihoods refers to payoff are maximal only for optimal candidate points by using Grey Level Co-occurrence Matrix (GLCM) [32] and data analysis [31]. The detailed survey presented that the suitable method required to improve the segmentation in retinal image scanning. TABLE 1 INFORMATION ABOUT DIFFERENT IMAGE SEGMENTATION TECHNIQUES Techniques Author & Ref Year Unsupervised segmentation Savelonas [8] 2011 Convex energy function Wu and Yang [9] 2012 Constrained Active contour Anh et al [7] 2012 Histogram Thresholding FCM Adaptive FCM Tan et al [10] 2011 Cao et al [11] 2012 Kernel measure+ FCM Gong et al [12] 2013 Graph based KMeans clustering Galluccio et al [13] 2012 Advanced K means Automated 2 Dimensional K means (A2DKM) Selvakuma r et al [14] Yusoff et al [15] 2012 Local clustering criterion Li et al [18] 2011 Non-linear probabilistic method Discriminative classification Unified level set model Prisacariu et al [20] 2011 Liu and Yu [19] Li et al [17] 2012 Multiphasemultichannel Kim and shan [16] 2011 Optical Disk (OD) localization and segmentation Yu et al [21] 2012 Atlas based approaches Merida et al [23] 2012 Multiscale graph cut approach Mahapatra et al [25] 2012 MFLAAM Toth et al [26] 2012 Gaussian Mixture Model Balafar [22] 2012 Maximum-APosteriori (MAP) Zhang et al [27] 2013 2012 2012 Performance Active contour based image segmentation A novel active contour based model segments the protein spots in two dimensional images with critical issues such as noise, streaks and multiplets. A convex optimization function with local Gaussian distributing fitting term with spatially variations of means and variances presented and the energy function formulated. Boundary refinement tool presented in constrained active contour has the capability to produce the smooth and accurate boundary contour. Fuzzy clustering The histogram thresholding techniques extracts the uniform regions of an image and FCM used to evaluate the compactness of clustering of uniform regions. Classical FCM employs the gain field model to correct the intensity in homogeneities by microscope imaging system. Gain field also regulates the center of cluster Kernel distance measure and trade off fuzzy weight factor estimates the extent of neighboring pixels. The objective function incorporates the kernel distance measure to improve the robustness to noise. K-means clustering The application Prim’s algorithm and Lloyd algorithm constructs the Minimum Spanning Tree (MST) and generalized cluster centroids to determine the number of clusters and location of cluster centroids. It detects the range and shape of Tumor in brain images and allows the accurate detection, reproducible and less execution time. Local and spatial information includes in A2DKM to determine the number of clusters. The comparative analysis on memory consumption of AFKM and A2DKM provided. Level set based image segmentation Level set formulation utilizes the local criterion function that defines the partition energy for simultaneous image segmentation and analysis the performance with presence of intensity inhomogeneity Elliptic Fourier descriptors represents the shape of the image. Segmentation carried out on nonlinear minimization of an energy function in latent space. The elimination of local minima and prediction of snap close to object boundaries performed with the help of level set function. It integrates the image gradient, region competition and prior information for CT liver tumor segmentation. the object indication improved by fuzzy clustering technique The energy function minimization performed for each segment corresponds to roof plans. Based on the topological relations, the intersection of adjacent roof segments reconstructs the model. Template matching identifies the Optical Disk location. Morphological filtering techniques eliminates the blood vessels and bright regions other than the OD. Intensity based image segmentation Two atlas based models such as probabilistic model and multi atlas registration based models applied to breast MR dedicated strategies. The set of 27 manually segmented volumes are used for testing Markov Random Field (MRF) models employs the multi-graph cut approach to achieve the sub-pixel registration and computation time reduction It utilizes the accurate algorithm for identification of Image Derived Attributes (IDA) offers effective segmentation and incorporates the level set implementation to overcome limitation in the specification of landmark and location of object interest in image. Presents a review of Gaussian Mixture Model (GMM) based image segmentation. The review of traditional works related to GMM based brain images also discussed. The nonsmooth nonconvex minimization problem investigated by MAP principle with relaxation in constraints of characteristic functions of partition regions Images Quality measurement Protein spots Volumetric overlap Volumetric error Synthetic and real images Influenze of weight function\ Segmentation level Accuracy Speed Buddhist image Home Girl Capsicum M-Fish dataset Brain image Salt & pepper corrupted image Algorithm Efficiency Intensity value Correct detection rate False detection rate Entropy based evaluation function Layout entropy Mars hyper spectral image Devis-Bouldin index Brain MR image Tumor shape Tumor position Execution time Memory consumption Lake House Hut Limon Vessel MR breast image Kinematics of person Accuracy CPU time Zebra Cheetah LTSC dataset Localized boundary Smoothness Volume overlap error Roof plan images Convergence rate MESSIDOR database Mean Absolute Distance, Correlation factor Breast MR image Dice coefficient Cine cardiac 3D liver perfusion Computational complexity T2 weighted prostate MRI Mean dice coefficient Accuracy T1 weighted MRI images Jaccard Similarity Index Dic similarity index Jaccard Similarity Coefficient CPU time Brain MRI Accuracy Detection filter Law and Chung[24] 2013 Gaussian distribution Sørensen et al [33] 2011 Mobahi et al [30] 2011 Texture descriptor Alvarez et al [28] 2012 Game theory Ibragimov et al [29] 2012 Linear filters Yuan et al [35] 2014 High fidelity models Tang et al [34] 2014 Grey Level Cooccurence Matrix (GLCM) Morphology and texture paradigm Reska et al [32] 2015 Noor et al [31] 2015 A novel intensity based algorithm segments the intracranial vessels and the attached aneurysms. Based on the topology structure, detected turbulent flows affects the low intensity regions. Texture based mage segmentation The textural dissimilarity between two Region of Interests (ROI) computed by k-NN classifiers and described by histogram of filter responses from the Gaussian filter bank Homogeneous texture region modelled as Gaussian distribution and the boundary of image described buy adaptive chain code. An agglomerative clustering process applied to image segmentation It employs convolutional neural network to recover the 3D scene layout from road side images with the features from noisy labels. A novel texture descriptor used to predict the maximal uniformity Application of dynamic program to optimal search problem between the landmarks in target image. Segmentation problem modelled as game theory and solution is equilibrium of candidate points, represents the landmark. Remote sensing image segmentation incorporates the spatial and texture information. The employment of linear filters enhances the spatial features. The weight index describes the segment ownership of pixels An object recognition system with sensing and collaboration information presented. The creation of high fidelity object models and the utilization in accurate detection and estimation provided. The integration of texture feature analysis with model evaluation analysis presented as two dimensional deformable model that uses edge and intensity based features. Thresholding and morphological based segmentation module coupled with feedback detects large deviations. The feedback model allows the detection of abnormal lung disease. III. SURVEY DISCUSSION Various techniques for image segmentation techniques are depicted. The results of the survey are shown in Table 1. The review of image segmentation techniques and classification emphasized that hierarchical framework in image segmentation techniques and how they are used to improve the quality of recognition. Segmentation based on active contour models conveyed that refinement of boundary was an important process in segmentation. Grouping of pixels of an image led the development of fuzzy based and K- means clustering process. The kernel metric approaches in addition to clustering eliminated the need of prediction of the number of clusters. K-means processes applied in various real world images such as brain MRI images, tumor segmentation for maximum accuracy. The Texture Based Encoding Segmentation (TBES) models emphasized the reduction of coding length and the uniformity nature of maximum likelihood regions provide the way to use angular based texture pattern and intensity deviation matrix to improve the quality of segmentation in proposed work. Intensity difference Input Image Threshold extraction Cluster formation Encode Lung cancer image Histogram dissimilarity Rank correlation Probabilistic Random Index(PRI), Berkeley Segmentation Data set (BSD) LabelMe dataset CamVid dataset Lung Chest Heart venticulars GeoEye-I satellite images Household objects Brodatz texture database High Resolution Computed Tomography Average time for image Confusion matrix Mean boundary distance Area overlap coefficient Correctness of segmentation Accuracy Confusion matrix Precision Recall Texture orientation Similarity Jaccard Index Area Overlap Error NO Label formulation Weight calculation Pixel direction Discontinuity Homogeneity Ratio (DH) PROPOSED WORK IV. The proposed multilevel set segmentation based on the combination of intensity and texture with level set functions consists of various processes. In the intensity analysis, extraction of threshold computes the edge intensity variation between the pixel and neighborhood pixels in matrix form. In the texture pattern analysis, angular texture pattern selects window coefficient by varying the angles for accurate image analysis. Based on intensity difference and texture pattern, the weight function is formed. The pixel matching process followed weight formation analyzes the matching of the pattern edge with the threshold value of the deviation matrix until the condition ( ) is satisfied. The updated weight matrix computes the contour formation over an image. Finally, application of the label on the basis of layer separation provides the segmentation output. The comparative analysis between angular texture pattern and traditional level set method shows that the effectiveness of proposed segmentation. The flow diagram of proposed method as shown in fig. 1. Preprocessing Window formation Vascular phantom Pattern extraction Pixel matching Weight updation Check weight Wi < Wi-1 Contour formation YES Segmentation output Fig. 1 Flow diagram of proposed method V. CONCLUSION In this paper, an overview of various image segmentation is presented. From the survey, it is find out that intensity and texture based methods based on level set function efficiently segment the image. The quality of the image with the presence of noise analyzed and improved on texture based methods. The analysis of the performance parameters such as accuracy, confusion matrix parameters, Kappa’s coefficient on traditional methods showed the suitable methods were required for improvement. The effective refinement of boundary on Optical Disk (OD) in retinal images performed by the level set methods. [17] [18] [19] [20] [21] REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] H. Narkhede, "Review of image segmentation techniques," Int. J. Sci. Mod. Eng, vol. 1, p. 28, 2013. J. F. Khan, S. M. A. Bhuiyan, and R. R. Adhami, "Image Segmentation and Shape Analysis for Road-Sign Detection," IEEE Transactions on Intelligent Transportation Systems, vol. 12, pp. 83-96, 2011. S. S. Al-Amri and N. V. Kalyankar, "Image segmentation by using threshold techniques," arXiv preprint arXiv:1005.4020, 2010. T. Schoenemann, F. Kahl, S. Masnou, and D. Cremers, "A linear framework for region-based image segmentation and inpainting involving curvature penalization," International Journal of Computer Vision, vol. 99, pp. 53-68, 2012. Z. Ji, Y. Xia, Q. Sun, Q. Chen, D. Xia, and D. D. Feng, "Fuzzy local Gaussian mixture model for brain MR image segmentation," IEEE Transactions on Information Technology in Biomedicine, vol. 16, pp. 339-347, 2012. H. Zhuang, K.-S. Low, and W.-Y. Yau, "Multichannel pulsecoupled-neural-network-based color image segmentation for object detection," IEEE Transactions on Industrial Electronics, vol. 59, pp. 3299-3308, 2012. N. T. N. Anh, J. Cai, J. Zhang, and J. Zheng, "Constrained active contours for boundary refinement in interactive image segmentation," in IEEE International Symposium on Circuits and Systems (ISCAS), 2012 2012, pp. 870-873. M. A. Savelonas, E. A. Mylona, and D. Maroulis, "Unsupervised 2D gel electrophoresis image segmentation based on active contours," Pattern Recognition, vol. 45, pp. 720-731, 2012. B. Wu and Y. Yang, "Local-and global-statistics-based active contour model for image segmentation," Mathematical Problems in Engineering, vol. 2012, 2012. K. S. Tan and N. A. M. Isa, "Color image segmentation using histogram thresholding–Fuzzy C-means hybrid approach," Pattern Recognition, vol. 44, pp. 1-15, 2011. H. Cao, H.-W. Deng, and Y.-P. Wang, "Segmentation of M-FISH images for improved classification of chromosomes with an adaptive Fuzzy C-Means Clustering Algorithm," IEEE Transactions on Fuzzy Systems, vol. 20, pp. 1-8, 2012. M. Gong, Y. Liang, J. Shi, W. Ma, and J. Ma, "Fuzzy c-means clustering with local information and kernel metric for image segmentation," IEEE Transactions on Image Processing, vol. 22, pp. 573-584, 2013. L. Galluccio, O. Michel, P. Comon, and A. O. Hero, "Graph based k-means clustering," Signal Processing, vol. 92, pp. 1970-1984, 2012. J. Selvakumar, A. Lakshmi, and T. Arivoli, "Brain tumor segmentation and its area calculation in brain MR images using Kmean clustering and Fuzzy C-mean algorithm," in International Conference on Advances in Engineering, Science and Management (ICAESM), 2012 2012, pp. 186-190. I. A. Yusoff, N. A. M. Isa, and K. Hasikin, "Automated twodimensional K-means clustering algorithm for unsupervised image segmentation," Computers & Electrical Engineering, vol. 39, pp. 907-917, 2013. K. Kim and J. Shan, "Building roof modeling from airborne laser scanning data based on level set approach," ISPRS Journal of Photogrammetry and Remote Sensing, vol. 66, pp. 484-497, 2011. [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] B. N. Li, C. K. Chui, S. Chang, and S. H. Ong, "A new unified level set method for semi-automatic liver tumor segmentation on contrast-enhanced CT images," Expert Systems with Applications, vol. 39, pp. 9661-9668, 2012. C. Li, R. Huang, Z. Ding, J. C. Gatenby, D. N. Metaxas, and J. C. Gore, "A level set method for image segmentation in the presence of intensity inhomogeneities with application to MRI," IEEE Transactions on Image Processing, vol. 20, pp. 2007-2016, 2011. Y. Liu and Y. Yu, "Interactive image segmentation based on level sets of probabilities," IEEE Transactions on Visualization and Computer Graphics, vol. 18, pp. 202-213, 2012. V. A. Prisacariu and I. Reid, "Nonlinear shape manifolds as shape priors in level set segmentation and tracking," in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2011 2011, pp. 2185-2192. H. Yu, E. S. Barriga, C. Agurto, S. Echegaray, M. S. Pattichis, W. Bauman, et al., "Fast localization and segmentation of optic disk in retinal images using directional matched filtering and level sets," IEEE Transactions on Information Technology in Biomedicine, vol. 16, pp. 644-657, 2012. M. Balafar, "Gaussian mixture model based segmentation methods for brain MRI images," Artificial Intelligence Review, vol. 41, pp. 429-439, 2014. A. Gubern-Mérida, M. Kallenberg, R. Martí, and N. Karssemeijer, "Segmentation of the pectoral muscle in breast MRI using atlasbased approaches," in Medical Image Computing and ComputerAssisted Intervention–MICCAI 2012, ed: Springer, 2012, pp. 371378. M. W. Law and A. Chung, "Segmentation of intracranial vessels and aneurysms in phase contrast magnetic resonance angiography using multirange filters and local variances," IEEE Transactions onImage Processing, vol. 22, pp. 845-859, 2013. D. Mahapatra and Y. Sun, "Integrating segmentation information for improved MRF-based elastic image registration," IEEE Transactions on Image Processing, vol. 21, pp. 170-183, 2012. R. Toth and A. Madabhushi, "Multifeature landmark-free active appearance models: application to prostate MRI segmentation," IEEE Transactions on Medical Imaging, vol. 31, pp. 1638-1650, 2012. H. Zhang, X. Ye, and Y. Chen, "An efficient algorithm for multiphase image segmentation with intensity bias correction," IEEE Transactions on Image Processing, vol. 22, pp. 3842-3851, 2013. J. M. Alvarez, T. Gevers, Y. LeCun, and A. M. Lopez, "Road scene segmentation from a single image," in Computer Vision– ECCV 2012, ed: Springer, 2012, pp. 376-389. B. Ibragimov, B. Likar, F. Pernus, and T. Vrtovec, "A gametheoretic framework for landmark-based image segmentation," IEEE Transactions on Medical Imaging, vol. 31, pp. 1761-1776, 2012. H. Mobahi, S. R. Rao, A. Y. Yang, S. S. Sastry, and Y. Ma, "Segmentation of natural images by texture and boundary compression," International journal of computer vision, vol. 95, pp. 86-98, 2011. N. M. Noor, J. C. Than, O. M. Rijal, R. M. Kassim, A. Yunus, A. A. Zeki, et al., "Automatic Lung Segmentation Using Control Feedback System: Morphology and Texture Paradigm," Journal of medical systems, vol. 39, pp. 1-18, 2015. D. Reska, C. Boldak, and M. Kretowski, "A Texture-Based Energy for Active Contour Image Segmentation," in Image Processing & Communications Challenges 6, ed: Springer, 2015, pp. 187-194. L. Sørensen, M. Nielsen, P. Lo, H. Ashraf, J. H. Pedersen, and M. De Bruijne, "Texture-based analysis of COPD: a data-driven approach," IEEE Transactions on Medical Imaging, vol. 31, pp. 70-78, 2012. J. Tang, S. Miller, A. Singh, and P. Abbeel, "A textured object recognition pipeline for color and depth image data," in International Conference on Robotics and Automation (ICRA), 2012 IEEE 2012, pp. 3467-3474. J. Yuan, D. Wang, and R. Li, "Remote sensing image segmentation by combining spectral and texture features," IEEE Transactions on Geoscience and Remote Sensing, vol. 52, pp. 16-24, 2014.