Medical Image Edge Detection Based on Neuro

World Academy of Science, Engineering and Technology International Journal of Computer, Electrical, Automation, Control and Information Engineering Vol:10, No:1, 2016

Medical Image Edge Detection Based on Neuro-Fuzzy Approach J. Mehena, M. C. Adhikary 

International Science Index, Computer and Information Engineering Vol:10, No:1, 2016 waset.org/Publication/10004525

Abstract—Edge detection is one of the most important tasks in image processing. Medical image edge detection plays an important role in segmentation and object recognition of the human organs. It refers to the process of identifying and locating sharp discontinuities in medical images. In this paper, a neuro-fuzzy based approach is introduced to detect the edges for noisy medical images. This approach uses desired number of neuro-fuzzy subdetectors with a postprocessor for detecting the edges of medical images. The internal parameters of the approach are optimized by training pattern using artificial images. The performance of the approach is evaluated on different medical images and compared with popular edge detection algorithm. From the experimental results, it is clear that this approach has better performance than those of other competing edge detection algorithms for noisy medical images.

Keywords—Edge detection, neuro-fuzzy, image segmentation, artificial image, object recognition. I. INTRODUCTION

E

DGES in a digital image provide important information about the boundary or contour where an abrupt change occurs in some physical aspect such as gray level value of an image. Edge detection is a frequently performed operation in many image processing applications because it is usually the first operation that is performed before other image processing tasks such as object recognition, boundary detection, image registration and image segmentation [1]. Consequently, the success of these subsequent image processing tasks is strictly dependent on the performance of the edge detection operation. Medical image edge detection is an important work for object recognition of the human organs and it is an essential preprocessing step in medical image segmentation [2]. The work of the edge detection decides the result of the final processed image. Classical methods of edge detection involve convolving the image with an operator (a 2-D filter), which is constructed to be sensitive to large gradients in the image while returning values of zero in uniform regions. There are an extremely large number of edge detection operators available, each designed to be sensitive to certain types of edges. Variables involved in the selection of an edge detection operator include edge orientation, noisy environment and edge structure. There are two main types of edge detector: one is the first derivative based edge detection operator to detect image edges by computing the image gradient values, such as Roberts, Sobel and Prewitt operator; the other one is the J. Mehena, Research Scholar, and M.C. Adhikary, Professor, are with the Department of Applied Physics and Ballistic is with Fakir Mohan University, Balasore, Odisha, India (e-mail: [email protected], [email protected]).

International Scholarly and Scientific Research & Innovation 10(1) 2016

second derivative based edge detection operator to detect edges by taking second derivative such as LOG and Canny operator [3], [4]. The first order derivative like Sobel, Prewitt and Laplacian of Gaussian operator, but in theory they belong to the high pass filtering, which are not fit for noisy medical image edge detection because noise and edge belong to the scope of high frequency. In real world applications, medical images contain object boundaries and object shadows and noise. Therefore, they may be difficult to distinguish the exact edge from noisy medical images. These detectors are simple to implement but they are usually inaccurate and highly sensitive to noise. The second order detects edges by taking second derivative and is very sensitive to noise too. The Gaussian edge detectors reduce the undesirable negative effects of noise by smoothing the image before performing edge detection [18]. Hence, they exhibit much superior performance over other operators especially in noisy conditions. The Canny detector [5], which is a Gaussian edge detector, is one of the most popular edge detectors and it has been widely used in many applications. Although the Gaussian detectors exhibit relatively better performance, they are computationally much more complex than classical derivative based edge detectors. Furthermore, their performances quickly decrease as the density of the corrupting noise increases. Mathematical morphology is a new mathematical theory which can be used to process and analyse the medical images [6]-[8]. It provides an alternative approach to image processing based on shape concept stemmed from set theory, not on traditional mathematical modelling and analysis. In the mathematical morphology theory, images are treated as sets, and morphological transformations which derived from addition and subtraction are defined to extract features in images. As the performance of classic edge detectors degrades with noise, morphological edge detector has been studied. It is a better method for edge information detecting and noise filtering than differential operation, which is sensitive to noise. This method is a better compromise between noise smoothing and edge orientation, but the computation is more complex than general edge detection algorithms. Therefore, a novel edge detector that is capable of extracting edges from medical images corrupted by noise is highly desirable. Fuzzy logic based approach can deal with the ambiguity and uncertainty in image processing in better way as compared to the traditional approaches [9]-[14]. When fuzzy logic is used for edge detection gives better result compare to the classical approach. The classical techniques like Sobel, Prewitt, Roberts, Canny edge detector have limitations of

229

scholar.waset.org/1999.4/10004525



ussing fixed vallue of parameeters or threshhold. The natture of eddges is not coonstant due too which few eedges left by being deetected. Howeever, as the medical im mage gets com mplex, different local areas will neeed different threshold valuues to acccurately find the real edgess. In addition, the global threeshold vaalues are deteermined manuually through experiments in the traaditional methhod, which leaads complex iin calculationn when larrge number of o different m medical images needs to bee dealt wiith. Fuzzy loggic is a soft coomputing techhnique that proovides uss flexibility by allowing the values without anyy such restrictions. Fuzzzy logic hass IF-THEN ruules and has simple s strructure to impplement and aalso this logic is a logical ssystem whhich is an extension off multi valuued logic. It is a prropositional caalculus in whiich there are m more than twoo truth vaalues. By usiing fuzzy tecchniques edgees having diffferent thickness can also be ddetected [19]. Fuzzy loggic is coonceptually eaasy to undersstand, flexiblee and is tolerrant of im mprecise data. Fuzzy logic is to map ann input space to an ouutput space annd for doing this a list off if then stateements caalled rules aree evaluated inn parallel. Theese rules are useful beecause they usse variables aand adjectives that describee those vaariables. This approach woorks well in nnoisy conditioon and siggnificantly dettects more edgges as comparred to the tradditional appproaches, butt still the edgess are not clearrly identified. In this paper, we extend oour research oon neuro-fuzzyy (NF) appproach for ddetection of edges in meedical images. This appproach is suittable to deal w with uncertainnty in the proccess of exxtracting usefu ful informatioon from noisyy images. Thhe NF syystems combinne the ability of neural nettworks to learrn and the capability of fuzzy logic ssystems to moodel the uncerrtainty. Heence, NF systtems may be employed ass powerful toools for eddge detection provided thatt appropriate network topoologies annd training straategies are choosen. In the prroposed methood, the eddges in the noisy image aare directly deetermined by a NF neetwork withouut needing a pre-filtering of the noisy input im mage. The NF F system consists of a num mber of subdettectors annd a postproceessor. Each ssub-detector evaluates a diffferent pixel neighborhhood in the filttering window w. The propossed NF eddge detector iis tested on various mediical images hhaving different imagee properties and a also comppared with popular The proposed NF edge deetector eddge detection algorithm. T exxhibits muchh better perfformance thaan the com mpeting opperators and m may efficienttly be used fo for the detectiion of eddges in medicaal images corruupted by varioous noises. The rest of tthis research ppaper is organnized as follow ws. In Seection II, the proposed appproach for m medical imagee edge deetection is deescribed in details. Experiimental resultts and coomparison w with existing edge detecction methodds are prresented in S Section III. Finally, concluusions follow wed by references comee in Section IV V. II. PROPOSSED APPROACH H The proposeed approach uses a desirred number oof NF suubdetectors wiith a postproccessor. Fig. 1 shows the ggeneral strructure of the proposed NF edge detectioon operator. T The NF suubdetector useed in this worrk operates on the 3x3 wiindow. Eaach subdetectoor evaluates a different neiighborhood reelation


betw ween the centeer pixel of thee filtering winddow and two oof its neigghbors. The ccoefficients annd possible eddge directionss for 3x3 windows aree shown in Fig. 2. The highher the numbeer of NF sub-detectorss, the better thhe edge detecction performaance, but the higher thee computationaal cost.

F Fig. 1 The geneeral structure of the proposed NF N edge detectioon opeerator

F Fig. 2 Coefficiennts and possiblee edge directionns for 3x3 windoow

A A. Neuro-Fuzzyy (NF) Subdettectors T The proposed NF subdetecttor is a first-oorder Sugeno type fuzzzy inference system s with 3-inputs 3 and 1-output. Thee NF subddetector uses in this work aare identical too each other. E Each inpuut has 3 trianngular type membership m f functions andd the outpput has The N NF subdetectorr uses in this work are idenntical to eeach other. Sinnce the NF suubdetector has 3 inputs and each inpuut has 3 mem mbership funcctions, the rule base contaiins a totaal of 27 rules. Inn this paper, we use the ttriangular mem mbership funcction andd uses singletonn fuzzifier andd Sugeno infeerence system.. The inpuut membershipp functions aree generalized ttriangular typee. ,

,

(1)

Herre the parametters , and are constants that characteerize the shape of the membership functions. Thhe fuzzy ruless are applied to the innputs and theen the output can be foundd by calcculating the weighted avverage of thee individual rule outpputs. ∑ ∑

(2)

ruule and wheere, denotees the output of the iss the weighting factor of each rule is calculated by evaluatingg the mem mbership exprressions in thee antecedent oof the rule. Thhis is accoomplished byy first convertting the inpuut values to fuzzy f mem mbership vallues by utillizing the innput memberrship funcctions and tthen applyingg the and ooperator to tthese mem mbership valuues. The andd operator coorresponds too the mulltiplication of input memberrship values.

230




B. Post-Proceessor The NF subddetectors outpputs are fed too the postproccessor, work output. The whhich generattes the finaal NF netw poostprocessor aactually calcuulates the aveerage value oof the suubdetector outpputs and comppares this valuue with a threeshold. Thhe threshold vvalue is the haalf of the availlable dynamicc range foor the pixel luuminance valuues. For 8-bitt images, wheere the pixel values rannge between 0 and 255, thee threshold vaalue is s trruncate this vaalue to 1228. The outputts of the NF subdetectors 0 or 255. The vaalue obtained at the output oof the postproocessor is also the outpuut value of thee NF edge detector and reprresents the decision whhether the centtre pixel of thee filtering winddow is ann edge pixel orr not.

in F Fig. 4. The eexperimental results and ccomparison off the propposed approacch with variouus edge detecttion methods such as S Sobel, Canny,, Morphologiccal and Fuzzyy logic approaaches are shown in Fig. 5. f Fig. 5 thhat the perform mance of the S Sobel Itt is observed from operator is very poor. For all test medical images, its ouutput imaages are severeely degraded by noise, mosst noise pulses are incoorrectly deteccted as edgees. The cannny edge deteector dem monstrates a coonsiderably beetter performannce than the S Sobel deteector. It corrrectly detectss most of tthe noise puulses. How wever, the efffect of noise is still clearrly visible as real edges are signnificantly disstorted by the noise. The morrphological eddge detector rremoves mostt of the noise, but the edges are noot clearly idenntified. On thhe other hand, the fuzzzy logic approoach exhibit veery good deteection perform mance andd still the edges are not identtified properlyy and the propposed approach successsful detects moore fine edges in all test meddical imaages.

Fig. 3 Trainingg of the subdetecctors

Each NF suubdetector iss trained inddividually annd the paarameters of thhe edge detecttor are optimizzed by trainingg. Fig. 3 represents thee setup used for training oof the subdeteectors. Thhe parameterss of the NF subdetectors under traininng are iteeratively adjussted so that itts output convverges to the output off the ideal edgge detector whhich, by definnition, can corrrectly deetect the locattions of the eddge pixels of tthe image fedd to its inpput. The ideall edge detectoor is conceptuaal only and dooes not neecessarily exisst in reality. IIt is only the output of thee ideal eddge detector that is neceessary for traaining, and tthis is reppresented by the desired ttraining imagee. It is a blacck and whhite image annd its black piixels indicate the locations of the truue edges of thhe input traininng image. Hennce it represennts the ouutput of the iddeal edge detecctor for the inpput training im mages. Thhe parameterss of the subddetectors undeer training aree then tunned by Levvenberg Marquardt basedd back-propaagation leaarning algoritthm [15]-[17]], which miniimizes the learning errror. III. EXPERIM MENTAL RESU ULTS ments are maade to veriffy the In this secttion experim peerformance off neuro-fuzzy aapproach on ddifferent test m medical im mages. The im mages that aree corrupted byy impulse noiise are coonsidered heree. All test imaages are 8-bitt grey level im mages. Thhe noisy imagges used in the edge detectiion experimennts are obbtained by coorrupting the original test images by im mpulse nooise with 5% nnoisy density. The noisy tesst images are shown


(a)

(b)

Fig. 4 Test medical images: MR im mages on (a) arre the original im mage andd the images on (b) are the noissy image used inn the edge detecction experriments

IV. CON NCLUSIONS T This paper pressented a neuroo-fuzzy approaach to detect more m finee edges of nooisy medical images. In tthis approach,, the paraameters of the NF subbdetectors unnder training are iteraatively adjusteed so that its output conveerges to the ouutput of tthe ideal edgee detector whiich, by definittion, can correctly deteect the locatioons of the edgge pixels of thhe image fed tto its inpuut. The expeerimental ressults show thhat the propposed approach is moree efficient forr medical imaage denoisingg and edge detection thhan the usuallly edge detecttion methods such Sobel, Canny, Morphologiccal and Fuzzy logic approacches. as S From m the experim mental resultss it is clear that the propposed approach can be used for efficcient extractioon of fine edgees in meddical images ccorrupted by im mpulse noise.

231




Sobel

Canny

Morphological

Fuzzy Logic

Proposed

Fig. 5 Comparison of the proposed approach with Sobel, Canny, Morphological and Fuzzy logic approaches

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10]

[11] [12] [13] [14] [15] [16] [17]

M.M. Subashini, S.K. Sahoo., “Pulse coupled neural networks and its applications”, International Journal of Expert Systems with Applications, Vol.41, pp.3965-3974, 2014. M.I. Rajab, M.S. Woolfson, and S.P. Morgan, “Application of regionbased segmentation and neural network edge detection to skin lesions”, Computerized Medical Imaging and Graphics, Vol. 28, pp.61-68, 2004. Raman Maini and Himanshu Aggarwal, “Study and Comparison of Various Image Edge Detection Techniques”, International Journal of Image Processing (IJIP), Vol. 3, 20, pp.1-12, 2010. R. C. Gonzalez, R.E. Woods and S.L. Eddins, “Digital Image Processing Using MATLAB”, 2nd Edn., Mc Graw Hill, New Delhi, 2010. J. Canny, “A Computational Approach to Edge Detection”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 8, pp. 679-687, 1986. Jing Xiao-jun, Yu Nong, and Shang Yong, “Image Filtering Based on Mathematical Morphology and Visual Perception Principle”, Chinese Journal of Electronics, Vol. 13, pp. 612-616, 2004. Richard A P, “A New Algorithm for Image Noise Reduction Using Mathematical morphology”, IEEE Transaction on Image Processing, Vol. 4, pp. 554-568, 1995. J. Mehena, “Medical Image edge detection based on mathematical morphology”, International Journal of Computer and communication technology, Vol. 2, pp.45-48, 2011. Liang, L.R., Looney, C.G., “Competitive fuzzy edge detection”, Applied Soft Computing, Vol. 3, pp. 132-137, 2003. J. Mehena and M.C. Adhikary, “Medical Image edge detection based on soft computing approach”, International Journal of Innovative Research in computer and communication Engineering, Vol. 3, pp.6801-6807, 2015. Choi YS, Krishnapuram R., “A robust approach to image enhancement based on fuzzy logic”, IEEE Trans. Image Processing, Vol.6, pp.808825, 1997. Abdallah A. Alshennawy, and Ayman A. Aly., “Edge Detection in Digital Images Using Fuzzy Logic Technique”, World Academy of Science, Engineering and Technology, pp.178-186,2009. Pushpajit A. Khaire and Nileshsingh V. Thakur, “A Fuzzy Set Approach for Edge Detection”, International Journal of Image Processing, Vol.6, pp.403-412, 2012. Russo, F., “Edge Detection in Noisy Images Using Fuzzy Reasoning”, IEEE Trans. On Inst. and Meas., Vol.47, pp.802-808, 1998. M. T. Hagan and M. B. Menhaj, “Training feed forward networks with the Marquardt algorithm,” IEEE Trans. Neural Network., Vol. 5, pp. 989-993,1994. Yuksel ME., “Edge detection in noisy images by neuro-fuzzy processing”, AEU International Journal of Communication, Vol. 61, pp.82-89, 2007. J. Vasavada and S. Tiwari., “An Edge detection method for grayscale images based on BP feedforward Neural network”, International Journal of Computer applications, Vol.67, pp.23-28, 2013.


[18] Hassanpour, H. and Asadi, S., “Image quality enhancement using pixel wise gamma correction”, International Journal of EngineeringTransactions B: Applications, Vol. 24, No. 4, pp. 301-311, 2011. [19] J. Mehena and M. C. Adhikary, “Medical Image Segmentation and Detection of MR Images Based on Spatial Multiple- Kernel Fuzzy CMeans Algorithm”, International Journal of Medical, Health, Biomedical, Bioengineering and Pharmaceutical Engineering, World Academy of Science, Engineering and Technology, Vol. 9, Issue 6, pp. 508 - 512, June-2015.

232
