An Algorithm for Coastline Extractionfrom Satellite Imagery

DEJAN VUKADINOVJohn Naisbitt University

Grad. School of Comp. Sci.Bul. umetnosti 29, Belgrade

SERBIAdejanvukadinov@gmail.com

RAKA JOVANOVICHamad Bin Khalifa University

Qatar Environment and Energy Research InstitutePO Box 5825, Doha

QATARrjovanovic@qf.org.qa

MILAN TUBA ∗

John Naisbitt UniversityGrad. School of Comp. Sci.Bul. umetnosti 29, Belgrade

SERBIAtuba@ieee.org

Abstract: Monitoring of coastline areas facilitates landscape development, sea transport, sea-level rise, changes incoastal areas and other important activities so it is very important that coastline extraction is quick and precise.In the past coastline extraction consisted of two stages: high-resolution images were captured from airplane andthen coastline was drawn based on these images. This operation was slow compared to today’s modern techniquessuch as satellite images and image processing. In this paper we proposed a new technique for coastline extractingfrom satellite images. The proposed procedure is based on algorithms for image processing and edge detection.Experimental results showed that the proposed method was fast and accurate.

Key–Words: Coastline extraction, histogram matching, Gaussian filter, locally adaptive thresholding, Canny edgedetector.

1 IntroductionCoastline extraction is technique based on analyzingsatellite images. These images are stored in adigital format. Each digital image is made up of aseries of pixels, and the pixels have attributes suchas color, brightness and position. Using computeralgorithms that can analyze the image pixels and theirmutual relations we can establish some rules for thefurther coastline extracting. The application of thesealgorithms on digital image is called digital imageprocessing techniques. Satellite images are a valuablesource of information, and, when comes to coastlineextraction, could be used for several purposes.

In the last 100 years human race activity heavilyinfluenced and threatened the Earth’s ecosystem. Thishas led to various problems for our environment.Nowadays, vegetation mapping is much easierbecause of capability of hyperspectral sensor todistinguish vegetation compared to other types ofterrain. Information on vegetation and wetlands canprovide a significant information about changes inhydrology and insight into impact of this changes onplant and animal world [1].

Satellite images analysis and information aboutcomposition of soil and climatic condition are useful

∗This research is supported by the Ministry of Education,Science and Technological Development of Republic of Serbia,Grant No. III-44006.

tool for agricultural crop yield prediction. Thesepredictions are very important because they providethe insight into the possible shortage of someagricultural crops which could affect national foodsecurity. In the past, this assessment was based onvisual assessment and samples taken directly from thefields, but this method was very slow and expensiveand is usually was used on smaller areas [2].

Coastline is an important basis for themeasurement of water and land resources. Thismeans that the fast and accurate coastline mappingis very important for the development of a country,urban planning and safe navigation [3].

Coastline extraction techniques can be classifiedinto three categories: image rectification andrestoration, image enhancement, informationextraction. Image rectification is the process ofcorrecting image geometry so it can be displayedon a flat surface. Image enhancement techniquesshould improve the quality of image to the level thatalow necessary information to be easily discerned inrelation to the sum. There are many different imageenhancement techniques. In the computer era, thesetechniques are mainly digital and algorithms areexecuted by computers. These digital techniques aremuch more precise and give much better results thanpreviously used image processing techniques.

For coastline extraction many different imageprocessing algorithms are combined. Numerous

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 35 Volume 5, 2017

adjustments and combinations were proposed in thepast several years. Some of the techniques thatare used are thresholding [4], contrast enhancement[5] usually by histogram equalization [6] and manyothers. The objective of the information extractionoperations is to replace visual analysis of the imagedata with quantitative techniques for automating theidentification of features in a scene. Aim of coastlineextraction is to classify all the pixels into the twogroups, water or land, usually by using some ofthe classification methods [7]. Some of classifierthat were used for classification in applications forcoastline extraction are support vector machine [8],Gaussian process [9], etc.

Another common task in applications forcoastline extraction is edge detection. One of thesuccessful method for edge detection is by usingfuzzy logic. Fuzzy image processing is consists ofall those approaches that understand, represent andprocess the images, their segments and features asfuzzy sets. The representation and processing dependon the selected fuzzy technique and on the problem tobe solved [10].

This paper describes the basic technologicalaspects of coastline extraction using satellite images.The rest of this paper is organized as follows. InSection 2 we will discuss articles about coastlineextraction, image processing and edge detection usingsatellite images. In Section 3 our proposed algorithmis presented. In Section 4 experimental results wereshowed and at the end in Section 5 a conclusion isgiven.

2 Literature ReviewIn [3], coastline detection methods were implementedto detect changes coastline of the Pearl River deltaarea in Guangdong province. One of the biggestproblems is caused by changes in coastal lines inthe Pearl River Estuary is narrowing of the riverwhich led to flooding in the top part of the riverbed. Their study provides an analysis of river Gorgeusing satellite images of Pearl River deltaic area inGuangdong Province, China. Methodology consistsof the following steps: a) data collection, b) imagedata processing and 3) visual and digital analysisof the coastline changes in time. It was concludedthat the better results can be obtained when animage separates into smaller parts, especially wherewetlands are. All satellite images must be with noclouds otherwise it will not be able to do correctextraction of the coastline. Based on the analysis ofimages each pixel has been distributed to some of thefollowing classes: water, forest, plantation and bare

land.In [4], an automated detection shoreline

method using histogram equalization andadaptive thresholding techniques is developed.Their study adopted Modified Self-AdaptiveHistogram Equalization with Plateau Methodthreshold (Modified Sapha-M), a clipped histogramequalization based contrast enhancement methodto enhance coastal features. Modified Sapha-M,is a modified method of Self-Adaptive HistogramEqualization Plateau (Sapho) proposed by Wang etal., 2006, to enhance the main objects and suppressthe background for infrared images. ModifiedSapha-M, which consists of five steps (Raju et al.,2013b); 1. Smoothen the input image histogramwith 3-neighbor Median filter 2. Fond the maximumlocal and global maximum values 3. Selected theoptimal mean plateau value 4. Modified the histogramaccording to mean plateau value and equalize the5. Normalized histogram the image brightness wasapplied. Thresholding operation was adopted wherepixels are coded with 0 (land pixels) and with 1(water pixels). Grouping of the region was done using’Grass fire’ method. This method works by takingfirst pixel on Picture as the beginning of the spread offire, as the fire spread by one grouped the pixels of thesame type by making one continuous area. For edgedetection Roberts edge operator by Thieler et al. andoutlined shorelines were converted to vector maps.

In [9], and method of extraction on coastlineremote sensing image is presented. This methoduses a Gaussian process classification for groupingland and water, and then extracting coastline from theobtained results. The results showed that the use ofGaussian process classification overcomes a series ofpublic problems about coastline extraction that over-learning and the limitations of poor generalizationability and small sample exist and Artificial NeuralNetwork and hyper-parameters are difficult to choosefor Support Vector Machine. This method is currentlyvery popular in the field of satellite analysis imagesand extracting data.

In [11], a coastline extraction method ispresented. It consist of a sequence of imageprocessing algorithms, in which the key componentis image segmentation based on a locally adaptivethresholding technique. Several technical innovationshave been made to improve the accuracy andefficiency for determining the land/water boundaries.They use of the Levenberg-Marquardt method and theCanny edge detector to speed up the convergence ofiterative Gaussian curve fitting process and improvethe accuracy of the bimodal Gaussian parameters.The result is increased reliability of local thresholdsfor image segmentation. A series of further image

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 36 Volume 5, 2017

processing steps are applied to the segmented images.Particularly, grouping and labeling contiguous imageregions into individual image objects enables themto utilize human heuristic knowledge about the sizeand continuity of the land and ocean masses todiscriminate the true coastline from other objectboundaries. The final product of their processingchain is a vector based line coverage of the coastline,which can be readily incorporated into a GISdatabase. Their coastline extraction method consistsof three groups of image processing algorithms: pre-segmentation, segmentation and post-segmentation.The pre-segmentation processing algorithms aim tosuppress image noise and enhance edge elements.

In [12] an approach is proposed based onfuzzy connectivity concepts and takes into accountthe coherence measure extracted from an InSAR(Interferometric Synthetic Aperture Radar) couple.The approach proposed is a semi-interactivesegmentation based on a seed-growing process.User is choosing where to plant a seed and then aseed-growing algorithm occupies neighbours landareas. The final result is a large land area whichhas all their pixels connected to the area where seedgrowing originally started.

A [13] presents a novel approach for coastlineextraction from SAR images. This method isusing the combination of modified K-means methodand adaptive object-based region-merging mechanism(MKAORM). For noise and speckle removing aregular and six offset Gaussian filters are applied.Along with Gaussian a 8x8 median filter is also used.Finally a modified K-means method is used to preparean image to land-water classification. For mergingmultiple land areas in one big area, a seed techniqueis used. Seed algorithm starts with planting a seedin one area selected by user and then allowing it togrow across the land using seed algorithm. The resultsshowed that 93 percent of detected coastline are in 2pixels range from each other.

In [14] a new methodology for automaticextraction of the coastline using aerial images waspresented. This is a combination of four stepalgorithm that is used to extract coastline. First anoise reduction filter is applied for the preparationof image for the next image processing step. Thenimage is separated into land and water regions usingthreshold technique which supply a binary image asa result. After that image is manually processed inorder to remove small objects and other irregularitiesso that at the end of the process only land and waterpixel remains. Finally an edge detection technique isapplied.

In [6] a procedure based on classifying theland and water components by applying a spectral

band ratios method is proposed. First step island-water classification based on the NormalizedDifference Vegetation Index (NDVI) and NormalizedDifference Water Index (NDWI). Second step isbinary conversion of the image by assigning 0 or 1values to land or water pixels. Third step is edgedetection using Sobel edge operator which is usinga pair of (3x3) kernels. Finally given edges areconverted to vector list. Experimental results showsthat the given algorithm is very fast and is very littleCPU consuming, and it is also very precise. Finalresults revealed the changes on Dubai coastline whichare mostly due to urbanization for touristic purposes.

In [15] an automated object-based regiongrowing integrating edge detection (OBRGIE) for theextraction of coastline was presented. Aquaculturecoast is special type of complex artificial coast. Itis formed by the farming of aquatic organisms ontidal flats. In recent years there is a rapid growth ofcoastal aquaculture, especially in some developingcountries. It has been estimated that aquaculturecoasts constitute about 30 percent of all coastlinesin mainland China. This is an enormous area thatneeds management and planning. So the first step iscoastline and aquaculture mapping. OBRGIE methodwas proved itself as a robust when it is applied onsmall and large areas. This method is very fast andcheap for coastline mapping using satellite images.OBRGIE method consists of four main steps. Firststep splits image to segments using multi resolutionsegmentation algorithm. This algorithm is verypopular in many applications. Smaller images allowsus to easier detect edges. Second step applies Cannyedge detection algorithm. This algorithm is appliedon a near-infrared (NIR) bands (band-3 for SPOT-5and band-5 for TM) as the edges of land and waterare sharp in these bands. Final step is post-processingwhich is removing unwanted objects and anomalies.Unwanted objects can be ships or lakes which isdetected as water areas (but if they are surroundedby the land they will be automatically replaced withland pixels). Resulting edge is usually jagged sothe special algorithm is applied for smoothing theedges and to get a natural look of coastline. Finallycoastline is converted to vector list using Douglas-Peucker algorithm. OBRGIE method has successfullycaptured land-water border, even on places wherecoastline is complex shape. Also OBRGIE methodshows that it is not dependable on image scaling, so itis successfully applied on small and large areas. Buton extremely large and small areas OBRGIE methodis not usable.

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 37 Volume 5, 2017

3 The Proposed AlgorithmThe aim of this paper is to develop a method toeffectively detect coastline. This approach producesvector files of the coastline which can be utilizedto estimate rates of change over relatively long timeperiod. In this study, various image processing andedge detection techniques were carried. It consists offive main steps:

1. Image Extraction

2. Histogram Matching

3. Gaussian Blur Filter

4. Locally Adaptive Threshold

5. Canny Edge Detection

3.1 Image Extraction

The first step is to select a desired area and extractit from satellite images. The dataset used in thepresent work for the extraction of the coastline iscomposed of SAR images acquired by Envisat, ERS,Landsat, IKONOS, DMC, ALOS, SPOT, Kompsat,Proba, IRS, SCISAT. All images are then scaled toproper resolution and converted to format needed forour program to start extracting coastline. Becausethe images from different satellite sources differsin illumination, contrast and type of day andweather conditions while taking this images, differentparameters and techniques are used in preparation theimages for the coastline extraction.

3.2 Histogram Matching

The next step is to apply histogram matchingtechnique. Histogram equalization represents imageprocessing technique is contrast based enhancementmethod and it is used to enhance coastal featuresbut it is also used in other applications [16], [17].By applying histogram equalization in most cases itwill increase image details, especially in places whereedge detection algorithm will be used. Instead ofusing standard histogram equalization a histogrammatching method is used. Standard histogramequalization is method where histogram is uniformlydistributed. Histogram matching technique producesmore details in coastline area, thus increasing theaccuracy of coastline detection.

3.3 Gaussian Blur Filter

Third step is to apply Gaussian blur filter. Gaussianblur filter (also known as Gaussian smoothing) is

the result of using Gaussian function to create blureffect on image. It is a very popular effect mostlyused in graphics software, typically to reduce noiseand speckle on images and to reduce detail. Theresult of this technique is a smooth blur. Gaussiansmoothing is also used as a pre-processing stagein computer vision algorithms in order to enhanceimage structures at different scales. In our experimentGaussian filter is used for noise and speckle removal,without blurring the major edge features. Gaussianfunction is 2D version of normal distribution (whichis one dimensional).

G(x, y) =1

2πσ2e−

x2+y2

2σ2

By applying normal distribution in 2 dimensions (xand y) we can calculate the weight matrix (Gaussiankernel also called Gaussian mask). There are twostandard sizes of the kernel: 3x3 and 5x5 pixels. The5x5 size creates more blur effect thus more noise isremoved. Then we must choose sigma, usually thegreater sigma removes more noise, in our case weused sigma with value 6. After calculating Gaussianmask, it is then applied to pixel values.

3.4 Locally Adaptive Threshold

Fourth step is to apply threshold technique.Thresholding is common technique for imagesegmentation and many different algorithms wereproposed for optimal threshold value determination[18], [19], [20], [21]. Instead of applying thresholdon a whole image, a locally adaptive thresholdingalgorithm is used for image segmentation. By usingthis technique we can increase coastline detectionby increasing edge visibility on areas in the imagewith low contrast. This technique will dynamicallyset the threshold according to the local characteristicsof neighbours pixels. The purpose of threshold is toseparate image into two main homogeneous regionswater and land. Then the border between land andwater regions can be delineated as the coastlines.In standard threshold technique applied on imageswith strong illumination gradient poor results areproduces. The larger image, the poorer results areproduced. We were using locally adaptive thresholdon smaller portions of the images from 7x7 pixels sizeto 20x20 pixels size. Also by using median instead ofmean the results were even better.

3.5 Canny Edge Detection

Edge detection using Canny algorithm. This verypopular algorithm is simple but reliable edge detection

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 38 Volume 5, 2017

Figure 1: Coastline extraction when histogrammatching was turned off (left) and with histogram(right)

Figure 2: Coastline extraction with difference in noiselevel: without blur (left) and with applied blur (right)

technique. Canny edge detection is a technique toextract edges between different vision objects. It hasbeen widely applied as satellite image edge detectiontechnique. Before using Canny edge detection, it ismandatory to apply Gaussian filter and thresholdingtechnique. Depending on what parameters are used inthese techniques it will greatly affect the Canny edgeresults.

4 Experimental ResultsThe proposed method has been tested with 2 satelliteimages representing different artificial islands ofDubai coastline, Palm Jebel and Palm Jumeirah. InFigure 4. you can see the image transformationwhen the techniques are applied. Images in first roware original images extracted from satellite imagery.In second row Histogram matching technique isapplied. Image contrast and color are affected by thistechnique. Edges between land and water are nowmore pronounced. Sometimes applying histogramwill degrade results (Figure 1.), it all depends on whatis the quality of the satellite images and were theyalready preprocessed sing some technique or filter.

In the third row are the images with appliedGaussian blur filter. After applying this filter there are

Figure 3: Coastline extraction by different thresholdvalues

much less noise and speckles on the images (Figure2.). Increasing the kernel size and sigma will increasethe blur effect on images. We must be careful not toexceed the blur, because we can loose fine details onthe image.

In forth row are the images with applied LocallyAdaptive Threshold. Threshold is an operation thatwill greatly affect the end results, so applying the rightthreshold value is critical. Higher threshold valueswill recognize shallow waters as the land area thusincrease the land, but reducing the threshold value toomuch will increase the water area on the expense ofthe land (Figure 3.).

Finally, a Canny Edge detection algorithmis applied which successfully detects coastline.Resulting coastline from Canny edge detection we canconvert to a vector list where coordinates of pixelsbecame vector coordinates. Although these imagesare very complex, obtained results are very good andaccurate.

5 ConclusionAlthough the coastline mapping seems to be a simpleoperation, in practice an automated extraction ofthe land/water is more difficult than one wouldexpect. Because the land/water boundaries onsatellite images lack of sufficient intensity contrastand the complications of distinguishing coastlinefrom other object like buildings etc, the process ofcoastline mapping is challenging even for todaysmodern computers and algorithms. So there is aneed for the better edge detection algorithms andimage processing techniques in the future. Despitethe successful results of our method on satelliteimages, it should be noted that the success rate ofcoastline mapping is heavily dependent from goodquality satellite images. Satellite images must bewithout clouds or it will greatly affect the finalresults. Using satellite images with different bandsgreatly increase the edge detection success rate, and

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 39 Volume 5, 2017

Figure 4: Coastline extraction by our proposedalgorithm step by step

combining different bands on one area is a winningcombination. During processing satellite images wedetected some delineation errors. These errors canbe easily corrected by deleting or inserting partsof an image using graphic editing tools. In futureresearch we recommend to develop an algorithmwhich automatically detect image properties likecontrast, brightness and color levels, and adjustthem for creating better results. For developingthis algorithm, we propose machine learning methodwhich would learn itself using hundreds or eventhousands images of the same location. This methodwill learn the machine to adapt itself not to depend onquality of the image or what type of day or weathercondition are affecting the results. For the best resultswe recommend a cross comparison method, wheretwo different coastline extraction algorithms will bedeveloped. They will be using images in different

bands, and then use a cross-checking to find thedifference in the results.

Acknowledgments: This research is supported bythe Ministry of Education, Science and TechnologicalDevelopment of Republic of Serbia, Grant No. III-44006.

References:

[1] C. Zhang and Z. Xie, “Combining object-based texture measures with a neural networkfor vegetation mapping in the Everglades fromhyperspectral imagery,” Remote Sensing ofEnvironment, vol. 124, pp. 310–320, September2012.

[2] Y.-P. Wang, K.-W. Chang, R.-K. Chen, J.-C. Lo,and Y. Shen, “Large-area rice yield forecastingusing satellite imageries,” InternationalJournal of Applied Earth Observation andGeoinformation, vol. 12, no. 1, pp. 27–35, 2010.

[3] X. Li and M. Damen, “Coastline changedetection with satellite remote sensing forenvironmental management of the Pearl RiverEstuary, China,” Journal of Marine Systems,vol. 82, pp. 54–61, 2010.

[4] R. Aedla, Dwarakish, and V. Reddy, “Automaticshoreline detection and change detectionanalysis of Netravati-GurpurRivermouthusing histogram equalization and adaptivethresholding techniques,” Aquatic Procedia,vol. 4, pp. 563–570, 2015.

[5] H. Demirel, C. Ozcinar, and G. Anbarjafari,“Satellite image contrast enhancement usingdiscrete wavelet transform and singular valuedecomposition,” IEEE Geoscience and remotesensing letters, vol. 7, no. 2, pp. 333–337, 2009.

[6] S. AL-Mansoori and F. AL-Marzouqi,“Coastline extraction using satellite imageryand image processing techniques,” InternationalJournal of Current Engineering and Technology,vol. 6, pp. 1245–1251, August 2016.

[7] D. Dai and W. Yang, “Satellite imageclassification via two-layer sparse codingwith biased image representation,” IEEEGeoscience and remote sensing letters, vol. 8,no. 1, pp. 173–176, 2011.

[8] M. C. Alonso and J. A. Malpica, “Satelliteimagery classification with lidar data,”

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 40 Volume 5, 2017

International Archives of the Photogrammetry,Remote Sensing, and Spatial InformationScience, vol. 38, no. 8, pp. 730–735, 2010.

[9] X. Hu, Z. Qin, J. Wang, and G. Su,“Coastline extraction on remote sensingimage using Gaussian process classification,” inInternational Conference on ElectromechanicalControl Technology, and Transportation,pp. 392–395, 2015.

[10] A. Alshennawy and A. Aly, “Edge detectionin digital images using fuzzy logic technique,”International Journal of Computer, Electrical,Automation, Control, and InformationEngineering, vol. 3, no. 3, pp. 540–548,2009.

[11] H. Liu and K. C. Jezek, “Automated extractionof coastline from satellite imagery by integratingCanny edge detection and locally adaptivethresholding methods,” International journal ofremote sensing, vol. 25, pp. 937–958, March2004.

[12] S. Dellepiane, R. D. Laurentiis, and F. Giordano,“Coastline extraction from SAR images anda method for the evaluation of the coastlineprecision,” Pattern Recognition Letters, vol. 25,pp. 1461–1470, August 2004.

[13] Z. Liu, F. Li, N. Li, R. Wang, and H. Zhang,“A novel region-merging approach for coastlineextraction from Sentinel-1A IW Mode SARimagery,” IEEE Geoscience and remote sensingletters, vol. 13, pp. 324–328, January 2016.

[14] V. Paravolidakis, K. Moirogiorgou, L. Ragia,M. Zervakis, and C. Synolakis, “Coastlineextraction from aerial images based onedge detection,” ISPRS Annals of thePhotogrammetry, Remote Sensing and SpatialInformation Sciences, vol. III-8, pp. 153–158,July 2016.

[15] T. Zhang, X. Yang, S. Hu, and F. Su,“Extraction of coastline in aquaculture coastfrom multispectral remote sensing images:Object-based region growing integrating edgedetection,” Remote Sensing, vol. 5, pp. 4470–4487, September 2013.

[16] M. Jordanski, A. Arsic, and M. Tuba, “Dynamicrecursive subimage histogram equalizationalgorithm for image contrast enhancement,”in 23rd Telecommunications Forum Telfor,pp. 819–822, Nov 2015.

[17] M. Tuba, M. Jordanski, and A. Arsic, “Chapter4 - improved weighted thresholded histogramequalization algorithm for digital image contrastenhancement using the bat algorithm,” in Bio-Inspired Computation and Applications in ImageProcessing (X.-S. Yang and J. P. Papa, eds.),pp. 61–86, Academic Press, 2016.

[18] I. Brajevic and M. Tuba, “Cuckoo search andfirefly algorithm applied to multilevel imagethresholding,” in Cuckoo Search and FireflyAlgorithm: Theory and Applications (X.-S.Yang, ed.), pp. 115–139, Springer InternationalPublishing, 2014.

[19] M. Tuba, “Multilevel image thresholding bynature-inspired algorithms-a short review,” TheComputer Science Journal of Moldova, vol. 22,no. 3, pp. 318–338, 2014.

[20] A. Alihodzic and M. Tuba, “Improvedbat algorithm applied to multilevel imagethresholding,” The Scientific World Journal,vol. 2014, p. 16, 2014.

[21] M. Tuba, N. Bacanin, and A. Alihodzic,“Multilevel image thresholding by fireworksalgorithm,” in 25th International ConferenceRadioelektronika, pp. 326–330, April 2015.

WSEAS TRANSACTIONS on COMPUTER RESEARCH Dejan Vukadinov, Raka Jovanovic, Milan Tuba

E-ISSN: 2415-1521 41 Volume 5, 2017