Manasi Final Project

download Manasi Final Project

of 22

Transcript of Manasi Final Project

  • 8/14/2019 Manasi Final Project

    1/22

    Segmentation of ComplexSegmentation of ComplexImages using SOM forImages using SOM for

    ClusteringClustering

    Manasi DatarManasi Datar

    [email protected]@cc.usu.edu

    April 22, 2004April 22, 2004

    Final Project for CS 7650Final Project for CS 7650

    Instructor:Instructor: Dr. Heng-Da ChengDr. Heng-Da Cheng

  • 8/14/2019 Manasi Final Project

    2/22

    Presentation OutlinePresentation Outline

    Core Concepts Segmentation

    Segmentation and clustering

    Self-Organizing Map (SOM) Clustering and SOM

    Segmentation and SOM

    My Idea :)

    Segmentation of complex images using SOM for

    clustering Approach: not novel but applied to a new domain :)

    Application: pre-processing of images for knowledge

    discovery

  • 8/14/2019 Manasi Final Project

    3/22

    SegmentationSegmentation

    What is it?What is it?A process to distinguish meaningful objects in anA process to distinguish meaningful objects in an

    image from the backgroundimage from the background

    How is it done?How is it done? threshold techniquesthreshold techniques

    edge-based methodsedge-based methods region-based techniquesregion-based techniques

    connectivity-preserving relaxation-based methodsconnectivity-preserving relaxation-based methods

  • 8/14/2019 Manasi Final Project

    4/22

    Segmentation TechniquesSegmentation Techniques

    Threshold techniquesThreshold techniques Make decisions based on local pixel informationMake decisions based on local pixel information

    Are effective when the intensity levels of the objectsAre effective when the intensity levels of the objects

    fall squarely outside the range of levels in thefall squarely outside the range of levels in the

    backgroundbackground

    Spatial information is ignoredSpatial information is ignored

    Edge-based techniquesEdge-based techniques Center around contour detectionCenter around contour detection

    Weakness: connecting together broken contour linesWeakness: connecting together broken contour lines

    Prone to failure in the presence of blurringProne to failure in the presence of blurring

  • 8/14/2019 Manasi Final Project

    5/22

    Segmentation TechniquesSegmentation Techniques

    contdcontd

    Region-based TechniquesRegion-based Techniques Connected regions: grouping neighboring pixels of similarConnected regions: grouping neighboring pixels of similar

    intensity levelsintensity levels

    Region merging: adjacent regions are merged under someRegion merging: adjacent regions are merged under some

    criterioncriterion Overstringent criteria: fragmentationOverstringent criteria: fragmentation Lenient: blurred boundaries overlooked (overmerge)Lenient: blurred boundaries overlooked (overmerge)

    Connectivity-preserving Relaxation-based TechniquesConnectivity-preserving Relaxation-based Techniques

    Active contour modelActive contour model start with some initial boundary shape and iteratively modifystart with some initial boundary shape and iteratively modify

    according to some energy functionaccording to some energy function

    Getting caught in a local minimum is a riskGetting caught in a local minimum is a risk

  • 8/14/2019 Manasi Final Project

    6/22

    Segmentation andSegmentation and

    ClusteringClusteringClustering

    data set as a group of clusters:

    collections of data points that belongtogether

    Segmentation as clustering represent an image in terms of clusters

    of pixels that belong together

  • 8/14/2019 Manasi Final Project

    7/22

    Clustering AlgorithmsClustering Algorithms

    Make each point a separatecluster

    Until the clustering is satisfactory

    Merge the two clusterswith the smallest inter- clusterdistance

    End

    Agglomerative

    Construct a single cluster containingall points

    Until the clustering is satisfactory

    Split the cluster that yieldsthe two components with the

    largest inter-cluster distance

    End

    Divisive

  • 8/14/2019 Manasi Final Project

    8/22

    Clustering for SegmentationClustering for SegmentationK-means algorithmK-means algorithm

    A natural objective function can be obtained by assuming that we know there arekclusters, where kis known.

    Each cluster is assumed to have a center; we write the center of the ith cluster asci.The ith element to be clustered is described by a feature vector x

    i

    We now assume that elements are close to the center of their cluster, yielding theobjective function

    Note: if the allocation of points to clusters is known, it is easy to compute the bestcenter for each cluster

  • 8/14/2019 Manasi Final Project

    9/22

    Clustering for SegmentationClustering for Segmentation

    contdcontd

    K-means Algorithm Choose kdata points to act as cluster centers

    Until the cluster centers are unchanged Allocate each data point to cluster whose center is nearest

    Now ensure that every cluster has at least one data point{possible techniques for doing this include supplying empty clusters with a

    point chosen at random from points far from their cluster center}

    Replace the cluster centers with the mean of the elementsin their clusters

    End

  • 8/14/2019 Manasi Final Project

    10/22

  • 8/14/2019 Manasi Final Project

    11/22

    Structure of the SOMStructure of the SOM

    The SOM arranges feature vectors The SOM arranges feature vectorsaccording to their internal similarity,according to their internal similarity,creating a continuous topological mapcreating a continuous topological mapof the input space.of the input space.

    In this topological map, the vectors thatIn this topological map, the vectors thatare similar in the input space areare similar in the input space aregrouped together, or clustered.grouped together, or clustered.

    The map is a two dimensional grid of The map is a two dimensional grid ofprocessing elements, called neurons.processing elements, called neurons.

    Each processing element is connectedEach processing element is connectedto each element of the input layer andto each element of the input layer and

    to its neighboring processing elements.to its neighboring processing elements.

    The connections between the layers The connections between the layers(weights) represent the strength of the(weights) represent the strength of theconnection.connection.

  • 8/14/2019 Manasi Final Project

    12/22

    The training algorithm is based on competitive learningThe training algorithm is based on competitive learning only winning processing elements are allowed to adjust weights or learnonly winning processing elements are allowed to adjust weights or learn

    After presenting an input vector, the processing elements response, orAfter presenting an input vector, the processing elements response, oractivation, is calculated by multiplying the input vector by a set of weightsactivation, is calculated by multiplying the input vector by a set of weights

    The processing element with the highest activation is declared winnerThe processing element with the highest activation is declared winner

    The winning processing element and its neighbors are allowed to adjustThe winning processing element and its neighbors are allowed to adjusttheir weights, while the remaining ones are left unchangedtheir weights, while the remaining ones are left unchanged

    The neighborhood is the group of processing elements adjacent to theThe neighborhood is the group of processing elements adjacent to thewinning neuronwinning neuron During training the size of the neighborhood is reduced to stabilize theDuring training the size of the neighborhood is reduced to stabilize the

    effect of the input vectors in the topological mapeffect of the input vectors in the topological map

    The result of the training is a centroid (representative vector) for eachThe result of the training is a centroid (representative vector) for eachcluster that minimizes the quantization errorcluster that minimizes the quantization error The quantization error is defined as the square root of the sum of theThe quantization error is defined as the square root of the sum of the

    squared difference of individual input patterns and calculated centroidssquared difference of individual input patterns and calculated centroids

    Working of the SOMWorking of the SOM

  • 8/14/2019 Manasi Final Project

    13/22

    SOM DemoSOM Demo

    I meant to show this in Matlab, butI meant to show this in Matlab, but

    unfortunately; this laptop does notunfortunately; this laptop does not

    have it installed:(have it installed:(

    So, here is a Java version insteadSo, here is a Java version instead

    demodemo

    http://www.arch.usyd.edu.au/~rob/java/applets/neuro/SelfOrganisingMapDemo.htmlhttp://www.arch.usyd.edu.au/~rob/java/applets/neuro/SelfOrganisingMapDemo.htmlhttp://www.arch.usyd.edu.au/~rob/java/applets/neuro/SelfOrganisingMapDemo.html
  • 8/14/2019 Manasi Final Project

    14/22

    SOM for ClusteringSOM for Clustering

    Inputimag

    e

    Topological map

  • 8/14/2019 Manasi Final Project

    15/22

    SOM for clusteringSOM for clustering

    Properties of SOMProperties of SOM ClassificationClassification

    VisualizationVisualization

    Can be used as a clustering tool :)Can be used as a clustering tool :)

    Will help in gaining useful informationWill help in gaining useful informationform segmenting complex images.form segmenting complex images.

  • 8/14/2019 Manasi Final Project

    16/22

    My IdeaMy Idea

    Segmentation of Landsat-7 TM images usingSegmentation of Landsat-7 TM images usingSOMSOM

    LandsatLandsatinformationinformation images freely available onlineimages freely available online

    Main reference:Main reference:

    Automatic Segmentation of MagneticResonance Images (MRIs) of the Human Brainusing Self-Organizing Maps (SOMs) byEvangelou

    http://www.transparentworld.ru/landsat/eng/landsat.htmhttp://www.transparentworld.ru/landsat/eng/landsat.htmhttp://www.transparentworld.ru/landsat/eng/landsat.htmhttp://www.transparentworld.ru/landsat/eng/landsat.htmhttp://www.transparentworld.ru/landsat/eng/landsat.htmhttp://www.transparentworld.ru/landsat/eng/landsat.htm
  • 8/14/2019 Manasi Final Project

    17/22

    Proposed MechanismProposed MechanismLandsat 7 image

    (raw data)

    CorrectionsRectified

    Landsat 7 imageSOM

    ConventionalClassifier Classified image

    Classified imagePost

    processing

    Comparison and Assessment

  • 8/14/2019 Manasi Final Project

    18/22

    Some ResultsSome Results

    Original Image Segmented Image

  • 8/14/2019 Manasi Final Project

    19/22

    In ConclusionIn Conclusion

    ResultsResults Comparison based onComparison based on

    Time complexityTime complexity EffectivenessEffectiveness

    SOM will be helpful in identifying intensitySOM will be helpful in identifying intensity

    patterns that exist in image datapatterns that exist in image data

    Knowledge of such patterns will proveKnowledge of such patterns will proveuseful interpretation of such imagesuseful interpretation of such images

    R fR f

  • 8/14/2019 Manasi Final Project

    20/22

    ReferencesReferences Evangelou, I.E.: Automatic Segmentation of Magnetic Resonance ImagesEvangelou, I.E.: Automatic Segmentation of Magnetic Resonance Images

    (MRIs) of the Human Brain using Self-Organizing Maps (SOMs). MS Thesis,(MRIs) of the Human Brain using Self-Organizing Maps (SOMs). MS Thesis,Clarkson University, Potsdam, NY, USA (1999).Clarkson University, Potsdam, NY, USA (1999).

    Fayyad, U.M.; Piatetsky-Shapiro, G.; Smyth, P. and R. Uthurusamy, R., (Eds.).Fayyad, U.M.; Piatetsky-Shapiro, G.; Smyth, P. and R. Uthurusamy, R., (Eds.).Advances in Knowledge Discovery and Data Mining. AAAI Press, the MIT Press,Advances in Knowledge Discovery and Data Mining. AAAI Press, the MIT Press,CA, USA (1996).CA, USA (1996).

    Rangsanseri, Y; et al.: Multispectral Image Segmentation using ART1/ART2Rangsanseri, Y; et al.: Multispectral Image Segmentation using ART1/ART2Neural Networks. 22nd Asian Conference on Remote Sensing, SingaporeNeural Networks. 22nd Asian Conference on Remote Sensing, Singapore(2001)(2001)

    Solaiman, B; Mouchot, M.C.: A Comparative Study of Conventional and NeuralSolaiman, B; Mouchot, M.C.: A Comparative Study of Conventional and NeuralNetwork Classification of Multispectral Data. IGARSS94, Pasadena, USA. (1994)Network Classification of Multispectral Data. IGARSS94, Pasadena, USA. (1994)

    Vesanto, J.: Data Mining Techniques Based on the Self-Organizing Map. MScVesanto, J.: Data Mining Techniques Based on the Self-Organizing Map. MScThesis, Helsinki University of Technology, Espoo, Finland (1997)Thesis, Helsinki University of Technology, Espoo, Finland (1997)

    Goldberg, M.; Shlien, S.: A clustering scheme for multi-spectral images. IEEEGoldberg, M.; Shlien, S.: A clustering scheme for multi-spectral images. IEEETransactions on Systems, Man and Cybernetics SMC 8. (1978)Transactions on Systems, Man and Cybernetics SMC 8. (1978)

    Theiler, J.; Gisler, G.: A contiguity-enhanced k-means clustering algorithm forTheiler, J.; Gisler, G.: A contiguity-enhanced k-means clustering algorithm forunsupervised Multispectral image segmentation. Proc SPIE 3159. (1997)unsupervised Multispectral image segmentation. Proc SPIE 3159. (1997)

    Landsat project info:Landsat project info: http://http://geo.arc.nasa.gov/sge/landsat/landsat.htmlgeo.arc.nasa.gov/sge/landsat/landsat.html

    fR f

    http://geo.arc.nasa.gov/sge/landsat/landsat.htmlhttp://geo.arc.nasa.gov/sge/landsat/landsat.htmlhttp://geo.arc.nasa.gov/sge/landsat/landsat.htmlhttp://geo.arc.nasa.gov/sge/landsat/landsat.htmlhttp://geo.arc.nasa.gov/sge/landsat/landsat.htmlhttp://geo.arc.nasa.gov/sge/landsat/landsat.html
  • 8/14/2019 Manasi Final Project

    21/22

    ReferencesReferences Jain, A. K.; Dubes, R. C.: Algorithms for Clustering Data. Prentice-Hall Inc.,Jain, A. K.; Dubes, R. C.: Algorithms for Clustering Data. Prentice-Hall Inc.,

    Englewood Cliffs, NJ, USA. (1988)Englewood Cliffs, NJ, USA. (1988)

    Haykin, S.: Neural Networks A Comprehensive Foundation, Pearson EducationHaykin, S.: Neural Networks A Comprehensive Foundation, Pearson Education(Singapore) Pte. Ltd. (2001)(Singapore) Pte. Ltd. (2001)

    Brown, D. A.; Craw I.; Lewthwaite, J.: Towards Core Image Processing with Self-Brown, D. A.; Craw I.; Lewthwaite, J.: Towards Core Image Processing with Self-Organizing Maps. (2001)Organizing Maps. (2001)

    Su, M.: Clustering Analysis, Department of Computer Science and InformationSu, M.: Clustering Analysis, Department of Computer Science and Information

    Engineering, National Central University (2003) {available at:Engineering, National Central University (2003) {available at:http://selab.csie.ncu.edu.tw/~muchun/course/cluster/Clustering-crisp.pdfhttp://selab.csie.ncu.edu.tw/~muchun/course/cluster/Clustering-crisp.pdf}}

    Mather P.: Computer Processing of Remotely Sensed Images, John Wiley &Mather P.: Computer Processing of Remotely Sensed Images, John Wiley &Sons, Inc. New York, NY, USA (1999)Sons, Inc. New York, NY, USA (1999)

    Paola, J. D.; Schowenderdt, R. A.: A review and analysis of back propagationPaola, J. D.; Schowenderdt, R. A.: A review and analysis of back propagationneural networks for classification of remotely sensed multi-spectral imagery,neural networks for classification of remotely sensed multi-spectral imagery,International Journal of Remote Sensing. (1995)International Journal of Remote Sensing. (1995)

    Landsat data:Landsat data:http://landsat.gsfc.nasa.gov/data/Browse/Cities/L7_CitiesGallery.htmlhttp://landsat.gsfc.nasa.gov/data/Browse/Cities/L7_CitiesGallery.html

    SOM toolbox for Matlab:SOM toolbox for Matlab: http://http://www.cis.hut.fi/projects/somtoolboxwww.cis.hut.fi/projects/somtoolbox//

    http://selab.csie.ncu.edu.tw/~muchun/course/cluster/Clustering-crisp.pdfhttp://selab.csie.ncu.edu.tw/~muchun/course/cluster/Clustering-crisp.pdfhttp://landsat.gsfc.nasa.gov/data/Browse/Cities/L7_CitiesGallery.htmlhttp://landsat.gsfc.nasa.gov/data/Browse/Cities/L7_CitiesGallery.htmlhttp://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://www.cis.hut.fi/projects/somtoolbox/http://landsat.gsfc.nasa.gov/data/Browse/Cities/L7_CitiesGallery.htmlhttp://selab.csie.ncu.edu.tw/~muchun/course/cluster/Clustering-crisp.pdf
  • 8/14/2019 Manasi Final Project

    22/22

    !!! Thank you for your patience !!!

    Your comments and suggestions are valuable please send