Automatic Fuzzy Object-based Analysis of VHSR Images for Urban 2

Automatic Fuzzy Object-based Analysis of VHSR Images for Urban 2
of 14
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  Automatic fuzzy object-based analysis of VHSR images for urban objects extraction Imane Sebari a, ⇑ , Dong-Chen He b a Filière de Sciences Géomatiques et Ingénierie Topographique, IAV Hassan II, Rabat, Morocco b Centre d’applications et de recherches en télédétection (CARTEL), Université de Sherbrooke, Sherbrooke, Québec, Canada a r t i c l e i n f o  Article history: Received 15 September 2012Received in revised form 5 February 2013Accepted 5 February 2013Available online 25 March 2013 Keywords: Automatic object extractionObject Based Image Analysis (OBIA)Fuzzy rule baseVHSR satellite imagesUrban areas a b s t r a c t Wepresentanautomaticapproachforobjectextractionfromveryhighspatialresolution(VHSR)satelliteimages basedonObject-Based ImageAnalysis(OBIA). Theproposedsolutionrequires noinputdata otherthan the studied image. Not input parameters are required. First, an automatic non-parametric coopera-tive segmentation technique is applied to create object primitives. A fuzzy rule base is developed basedon the human knowledge used for image interpretation. The rules integrate spectral, textural, geometricand contextual object proprieties. The classes of interest are: tree, lawn, bare soil and water for naturalclasses; building, road, parking lot for man made classes. The fuzzy logic is integrated in our approach inorder to manage the complexity of the studied subject, to reason with imprecise knowledge and to giveinformation on the precision and certainty of the extracted objects. The proposed approach was appliedto extracts of Ikonos images of Sherbrooke city (Canada). An overall total extraction accuracy of 80% wasobserved. The correctness rates obtained for building, road and parking lot classes are of 81%, 75% and60%, respectively.  2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by ElsevierB.V. All rights reserved. 1. Introduction Mapping impervious surface from remote sensing imagery issignificant to a range of issues especially to sustainable develop-ment of urban areas (Demarchi et al., 2012; Gao et al., 2012;Longleyet al., 2005; Hollandet al., 2006;Donnayet al., 2001). Veryhigh spatial resolution (VHSR) images and advanced image pro-cessing algorithms both driven the technologic advance in remotesensing of impervious surfaces (Xu, 2013). One of the emergingtrends in this field is OBIA, object-based image analysis (Weng,2012). OBIA is considered as a powerful tool for classification andanalysis of VHSR images compared to the traditional per-pixelclassifiers (Blaschke, 2010; Navulur, 2007; Blaschke et al., 2000).The advantage of OBIA is that it does not use individual pixelsbut adjacent pixel groups that can be characterized by spectral,textural, geometric and contextual information. Taking this infor-mation into account through the object-based approach allowobtaining enhanced results (Campbell, 2007).Object based image analysis has been defined as a new disci-pline at the first international conference on Object-Based ImageAnalysis:‘‘ Object-Based Image Analysis (OBIA) is a sub-discipline of GIScience devoted to partitioning remote sensing (RS) imagery intomeaningful image-objects, and assessing their characteristicsthrough spatial, spectral and temporal scale. ’’ (Hay and Castilla,2006).The OBIA, also called object oriented image analysis, aims toreplicate and/or to surpass the human interpretation of imagesautomatically or semi-automatically (Hay and Castilla, 2006).Two main stages can form the OBIA process: (1) creation of imageobjects and (2) classification of image objects. Usually, the firststep is performed through a segmentation technique (Lang andBlaschke, 2006; Jensen, 2005). This step is a crucial since it pro-vides the basic units (image objects) on which later process willbe applied. Therefore, the success of OBIA approach is related tosegmentation quality. The second stage, classification, tries to cre-ate ‘real’ objects from ‘image’ objects. The classification method ischosen with relation to the desired goal, to the studied image, andalso to its ability to integrate ancillary information. Several meth-odscanbeusedatthetwostagesoftheOBIAapproach.Thechosenalgorithmsstronglyinfluencethefinal results(LuandWeng, 2007;Caloz and Pointet, 2003).The first known reference that used the object-based approachwas Kettig and Landgrebe (1976). They proposed a classificationapproach of multispectral images by extracting and classifyinghomogenous objects. Their approach consists first in subdividingthe image in spectrally homogenous pixel groups. These groupsare then classified through supervised technique (maximumlikeli-hood). They applied their approach to aerial and satellite images 0924-2716/$ - see front matter   2013 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS) Published by Elsevier B.V. All rights reserved. ⇑ Corresponding author. Tel.: +212 650632611; fax: +212 37680180. E-mail addresses:, (I. Sebari). ISPRS Journal of Photogrammetry and Remote Sensing 79 (2013) 171–184 Contents lists available at SciVerse ScienceDirect ISPRS Journal of Photogrammetry and Remote Sensing journal homepage: Downloaded from  (Landsat). This method is still used under the name of ECHOs(Extraction and Classification of Homogeneous Objects). It is avail-able in an open-source environment – Multispec (Biehl and Land-grebe, 2002). Lee et al. (2003) applied it to extract shapes and positions of building from Ikonos image. The ECHO approach hasbeen used by Jiménez et al. (2005) under a non-supervised version(UnECHO) to extract homogenous regions from hyperspectralimages.The use of OBIA became more widespread with the advent in2000 of eCognition, the first commercially available, object basedimage analysis software (Blaschke, 2010). It is today known asDefiniens. The object image creation step is conducted by a multi-resolution segmentation based on the Fractal Net Evolution Ap-proach (Baatz and Schäpe, 2000). The segmentation algorithm isa bottom-up region-growing technique. The growing decision isbased on local homogeneity criteria describing the similarity of adjacent image objects in terms of size, distance, texture, spectralsimilarityandform(BaatzandSchäpe, 2000). User-definedthresh-olds are interactively used to decide whether objects are mergedinto larger objects or not. For image object classification, twomethods are proposed: the nearest neighbor and fuzzy rules base(Benz et al., 2004). Several attributes (spectral, geometric and con-textual) can be integrated in the classification process. Significantstudies have used this software in different applications (forestry,urban, agriculture, coastal zone, etc.). Regarding VHSR satelliteimages of the urban area, they attempted to interpret either thewhole image (Myint et al., 2011; Kux and Araújo, 2008; Marchesi et al., 2006; Caprioli and Tarantino, 2003; Mittelberg, 2002; Kress-ler et al., 2001; Meinel et al., 2001), or to extract specific objectssuch as buildings (Hofmann, 2001a), roads (Repaka et al., 2004; Nobrega et al., 2008), and private gardens (Mathieu et al., 2007) or informal settlements (Hofmannet al., 2008). Recently, other im-age analysis software has developed OBIA modules like FeatureAnalyst (Tsai et al., 2011) or ENVI Feature Extraction (Hu and Weng, 2011).Comparedtothepixel-basedapproach, theextractedobjectsbyanOBIAapproachare morehomogeneousthan bypixels basedap-proach and are closer to a visual human interpretation (Huipinget al., 2003). The OBIA’s results (extracted objects) can be inte-grated within vector GIS more easily than classified raster maps(Walter, 2004). The OBIA can be applied on different satelliteimages. However, it has proven to be particularly appropriate forVHSR imagery especially in urban areas ( Jacquin et al., 2008;Campbell, 2007; Mo et al., 2007; Herold et al., 2003; Shackelfordand Davis, 2003; Rego and Koch, 2003; Bauer and Steinnocher,2001;Hofmann,2001b).TheclassificationofVHSRimagesofurbanenvironment can show some imprecision due to the nature of these images and the studied objects. In fact, satellite images’ pix-elscancorrespondtoseveralobjectsof differentnatures. Thismix-ture causes imprecision in the classification of these pixels. WithVHSR images, the problem is less important but still present. Also,somedifferentobjectscanyieldcloseor similarspectralresponses.This problem is greater in urban areas. Several objects of differentclasses yield the same spectral responses. This is due to the use of the same building materials and/or by the low spectral resolutionof the images (Herold et al., 2004).The researches on object extraction fromVHSR images in urbanareas which applied the OBIA approach are various. They differaccording to the objects to extract and also to the methods usedineachstage.Theextractioncanconcernonlyoneobjectclass(likebuilding (Lee et al., 2003), Roads (Repaka et al., 2004)) or try to interpret the whole image (Shackelford and Davis, 2003). The usedclassification methods can vary fromthe more conventional, usingonlyspectralinformation(Myintetal.,2011),tothemorecomplex,basedonexternal knowledge(Forestier et al., 2012; Bouziani et al.,2010). In OBIA approach, the integrated knowledge can be relatedto the objects’ characteristics (spectral, geometric and contextualproperties, relationships between objects, etc.), to information ini-tially relative to the extraction (object models, constraints, etc.) orto the used data (date and position of the sensor) (Baltsavias,2004). Adopting a classification method based on knowledgewouldgive the opportunity to take more informationintoaccount.This allows better discrimination between object classes and effi-cient extraction of objects (Campbell, 2007). Rule-based systemsbelongtoknowledge-basedmethods that simulatethehumanrea-soningmechanismandtranslateknowledgethroughdecisionrules(Tso and Mather, 2001). Fuzzy logic can also be integrated in clas-sification methods in order to resolve knowledge representationand classification problems in complex environments (Han et al.,2005). Thus, fuzzyrules consist of a set of fuzzy expressions allow-ing the evaluation of specific attributes. In comparison with a clas-sical rule, the response to a fuzzy rule is given with a degree thatexpresses the satisfaction of this rule, simultaneous applicationof several rules is allowed and an object can have different mem-bership degrees to the studied classes (Dubois et al., 2007). The fi-nal decision can be taken according to the rule to which themembership degree is maximal.Several studies have used fuzzy rule base in OBIA approach toextract urban object from VHSR. A review of these studies can befound in Weng (2012) and Blaschke (2010). They have shown high extraction precisions. However, during their process, values of parameters and thresholds were set manually. One of the recentOBIA research is directed towards the automation of image pro-cessing (Blaschke, 2010). This should concern both segmentationand classification methods used during OBIA process. For segmen-tation, automatic techniqueshouldallowthecreationof imageob- jects without setting any parameters or homogeneity threshold.For a fuzzy rule base, an automatic solution will allowthe general-izationof decisionrules on other images of different types withoutreformulating other rules more adapted to the new context. For-mulating rules based on human expert knowledge is not alwayseasy. Ascertaining thresholds and weights of rules is usually leftto the humanuser or based on training data (Walter, 2004; Ishibu-chi et al.,1992; Puissant et al., 2006).In this paper, we propose a methodology for automatic extrac-tion of urban objects from VHSR satellite images through an ob- ject-based image analysis approach. The proposed solutionrequires no input data other than the studied image. Not inputparameters are required. Segmentation is conduct by a nonpara-metric cooperative technique. The extraction is based on a fuzzyrule base adapted for interpreting VHSR images in urban areas.The extracted objects are organized in layers with information onthe precision and certainty of their extraction. This paper is orga-nized as follows. We describe first the proposed methodology.We present then the test data as well as obtained results. A lastsection will present the analysis and discussion of these results. 2. Proposed approach We proposed a new approach for automatic extraction of ob- jects from VHSR images. The objects of interest concern principalurban object and are presented in Table 1. The proposed approachadopts object-based image analysis principle and is constitutes of twoprincipal steps: creationof primitivesfrompixels andcreationof objects from primitives. The primitive corresponds to an inter-mediate state between the pixel and the object to extract. It con-sists of a group of homogenous adjacent pixels. The first step isconducted by segmentation technique and the second by fuzzyrule base. These two steps are performed automatically withoutneed to introduce parameters. The extracted objects are organizedinlayersbyclasses.Individuallayersareoverlaidtoproduceafinal 172  I. Sebari, D.-C. He/ISPRS Journal of Photogrammetry and Remote Sensing 79 (2013) 171–184  output map of extracted urban objects. Information on the qualityof theextractionis alsoprovided. Fig. 1presentstheconceptof ourapproach.  2.1. Creation of primitives from pixels Segmentation technique is used to create the primitives frompixels. The image segmentation algorithm used in this study fol-lows the approach given in Sebari and He (2009) where the seg-mentation requires no parameters and no input data other thanthe images to be processed. It is based on cooperation between re-gion-growing segmentation and edge segmentation. The segmen-tation adopts sequential region-edge cooperation. The edgesegmentation is performed first on panchromatic band and inte-grated into multispectral region growing segmentation as addi-tional criteria for seeds selection and for segmentation criteriadefinition. The approach uses a spectral homogeneity criterionwhose threshold is adaptive. It varies across the image dependingon the object to be segmented and its neighborhood. It is moreappropriate than a single threshold to apply to the entire imageespecially for complex images like VHSR images. The Fig. 2 pre-sents the principle of the adopted segmentation.The threshold of spectral homogeneity is calculated automati-cally for each new segment in every spectral band during the pro-cess of segmentation: Once the seed is chosen and beforeaggregating the pixels, a window of analysis is centered on it andthe spectral values of pixels contained in this window are consid-ered to elaborate a frequency histogram for differentiated values.For this purpose, a band differentiation algorithm is used to deter-mine the absolute maximum difference between a pixel and itsneighbors for each pixel: d v  ð i ;  j Þ ¼  max  1 6 k 6 1  1 6 l 6 1 j v  ð i ;  j Þ  v  ð i  þ  k ;  j  þ  l Þj ð 1 Þ With  dv ( i ,  j ) is the differentiated value at the pixel ( i ,  j ) and  v  ( i ,  j )is the value of pixel ( i ,  j ). In homogenous areas,  dv  has small values,whereas  dv  takes larger values in the boundaries between regions.The overall shape of the histogram is bimodal (Fig. 3): Since morepixels are inside objects than boundaries, the first peak corre-sponds to pixels inside the objects and the second represents thepixels at the boundaries. If   dv  is considered as the spectral homo-geneity factor  h , the threshold  T   is considered as the valley be-tween the two peaks. The automatic detection of this valley isdone according to a modified technique of  Zack et al. (1977). Thevalley corresponds to the point in the histogram that is farthestfrom the straight line joining the two peaks of the histogram.A pixel may be assigned to a segment if it satisfies in the nbands of the image the following condition: fð h 1  <  T  1 Þ AND . . . AND  ð h b  <  T  b Þ AND . . . AND ð h n  <  T  n Þg ; with ;  b  ¼ f 1 ;  . . .  ; n g ð 2 Þ Afterthesegmentation,theprimitivesaretransformedintovec-tor format and represented by polygons. Our choice of the vectormode is justified by many reasons:- Easy definition of geometric properties.- Explicit topology between the different objects to define con-textual properties.- Possibility to describe a segment through several attributesstored in a database.- Easier to integrate the extracted objects’ layers in existing geo-graphic database.- Possibilityof overlayingthepolygons’ layer withother layers inraster or vector format.- Easier to compare to others geographic database.  2.2. Creation of objects from primitives The step of passing from primitives to object is conducted byapplying a fuzzy rule base. This fuzzy rule base contains a knowl-edge used by a human photo-interpreter to identify urban objects.  Table 1 Objects of interest. Level I Level II Level IIINatural classes Vegetation LawnForestTreeBare soil Bare soilWater RiverLakeMan-made classes Road network RoadParking lotBuilding Building    E  x   t  r  a  c   t  e   d   i  n   f  o  r  m  a   t   i  o  n   P  r  o  p  o  s  e   d   A  u   t  o  m  a   t   i  c   O   b   j  e  c   t  e  x   t  r  a  c   t   i  o  n  a  p  p  r  o  a  c   h   I  n  p  u   t   d  a   t  a B G R NIRMultispectralsegmentation Fuzzy rule baseObject classes    M  e  m   b  e  r  s   h   i  p   d  e  g  r  e  e Vegetation Building Road Parking lotSegmented imagePANInformation onextraction quality Fig. 1.  Proposed automatic object extraction approach. I. Sebari, D.-C. He/ISPRS Journal of Photogrammetry and Remote Sensing 79 (2013) 171–184  173  The methodology adopted to establish the rules is as follow:knowledge modeling, fuzzy rule base creation and assignment of objects to classes.  2.2.1. Knowledge modeling  The purpose of this step is to model the knowledge that an ex-pert uses in order to identify urban objects. Five photo-interpreta-tion keys (Paine and Kiser, 2003) were used to define objects’properties: size, shape, color (spectral response), texture and sha-dow. We haveconsidered spectral, textural, geometric andcontex-tual properties to describe studied objects. Then, a quantifiableattribute is associated to each object property. A quantifiable attri-bute is defined by a mathematical formulation. The choice of attri-butes is based upon those used in the literature and validated bytests. The Table 2 presents the quantified attributes adopted foreach property.Then, we have done a discriminative analysis which purpose istodefine, foreachattribute,adiscriminativethresholduponwhichtheattributewillcharacterizetheassociatedproperty.Theanalysisconsists in studying the mathematic formulation and variation of each attribute. Afterwards, two approaches were followed todefine attribute’s discriminative thresholds depending on whetherthe studied attribute is dependent or not on the image. For theattributes that are independent on the image (like elongation in-dex, compactnessindex, etc.), thethresholdsaredeterminedbasedon their use in the literature and on tests. For the attributesdepending on the image, methodologies are proposed to automat-icallydefine the corresponding discriminative thresholds. Thecon-cerned attributes are the brightness index (shadow property) andarea (large property).The methodology used to automatically determine the bright-nessindexthresholdisbasedonthehistogramoffrequenciesofin-dex’s values in the studied image. The threshold is chosen as thefirstvalleyinthehistogram(Fig. 3). Thisisjustifiedbythefactthatshadow areas in VHSR image present low spectral values in thefour bands. We consider that, for an urban area image, and onthe frequency histogram of brightness index, the first peak reflectsshadow areas. It reflects also water or low albedo materials. But,since the objects will be described by multicriteria rules, the useof other attributes will allow the discrimination between shadowand the others objects.The automatic extraction of the brightness index threshold isperformed by applying the proposed algorithm: (1) Searchingautomatically first for the value corresponding with the first peak(first maximal value), (2) searching automatically thereafter forthe value of the threshold that satisfies the criteria: its brightnessindexvalueishighertotheoneofthefirstpeakanditiscomprisedwithin two values whose frequencies are higher than the fre-quency of this value.The area index is used to describe large size objects. Definingarea index threshold is based on the knowledge that, on an urbanVHSRimage, therearemoreobjectsofsmall andmediumsizethanlarge-size objects. Thus, the frequency of an image’s objects’ areavalues histogram looks like described in Fig. 4. The automaticextraction of this threshold is performed by using a modified ver-sion of the triangle technique proposed by Zack et al. (1977). It isperformed through an algorithm that searches for the value corre-sponding to the farthest point on the line joining the greatest peakand the last point in the histogram with the maximum area value.  Automatic determinationof spectral homogeneitythresholds for n  bands Add to the region'sboundary New segment  For each pixel in the list  YesNo SegmentLook for the 8unassignedneighboring pixels Add to the list of pixels to checkT 1 , …,T n  Add the pixel to theregion Seed  Check the spectral homogeneity criterion and the edge criterion ≠ Look for the 8 unassignedneighboring pixelsCheck the adjacency criterion h b  < T b  | b={1,…,n} edge pixel Fig. 2.  Adopted segmentation approach (Sebari and He, 2009). Differentiated values (dv) Threshold selected ObjectsEdges       P      i    x    e      l    s Fig. 3.  General shape of the histogram of the differentiated values.174  I. Sebari, D.-C. He/ISPRS Journal of Photogrammetry and Remote Sensing 79 (2013) 171–184


Jul 22, 2017


Jul 22, 2017
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks