Your SlideShare is downloading. ×
Hd2412771281
Hd2412771281
Hd2412771281
Hd2412771281
Hd2412771281
Upcoming SlideShare
Loading in...5
×

Thanks for flagging this SlideShare!

Oops! An error has occurred.

×
Saving this for later? Get the SlideShare app to save on your phone or tablet. Read anywhere, anytime – even offline.
Text the download link to your phone
Standard text messaging rates apply

Hd2412771281

117

Published on

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

IJERA (International journal of Engineering Research and Applications) is International online, ... peer reviewed journal. For more detail or submit your article, please visit www.ijera.com

Published in: Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total Views
117
On Slideshare
0
From Embeds
0
Number of Embeds
0
Actions
Shares
0
Downloads
3
Comments
0
Likes
0
Embeds 0
No embeds

Report content
Flagged as inappropriate Flag as inappropriate
Flag as inappropriate

Select your reason for flagging this presentation as inappropriate.

Cancel
No notes for slide

Transcript

  • 1. Sameena Banu / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.1277-1281 The Comparative Study on Color Image Segmentation Algorithms Sameena Banu, Assistant Professor (Khaja Bandanawaz College of Engineering,Gulbarga) Apparao Giduturi, HOD, CSE (Gitam University, Vishakhapatnam) Syed Abdul Sattar, HOD, E & CE (Royal Institute of Technology, Hyderabad)Abstract Image segmentation is the initial step in segmentation algorithms are given. Finally, lastimage analysis and pattern recognition. Image section provide conclusion.segmentation refers to partitioning an image intodifferent regions that are homogenous or similar in 2. COLOR SPACESsome characteristics. Color image segmentation Several color spaces such as RGB, HIS,attracts more and attention because color image CIE L* U* V* are utilized in color imagecolor images can provide more information than segmentation. Selecting the best color space still isgray level images. Basically, color image one of the difficulty in color image segmentation.segmentation approaches are based on monochrome According to the tristimulus theory color can besegmentation approaches operating in different color represented by three components, resulting fromspaces. This paper gives the comparative study of three separate color filters as follows:different color image segmentation algorithSamsand reviews some major color representationmethods. R = ∫λ E( λ) SR(λ) dλ G = ∫λ E( λ) SG(λ) dλKey word- Color image segmentation, color R = ∫λ E( λ) SB(SR) dλspaces, edge detection, thresholding. Where, SR ,SR ,SR are the color filters on the 1. INTRODUCTION incoming light and λ is the wavelength. Color image segmentation is a process ofextracting from the image domain one or more Form R,G,B representation, we can deriveconnected regions satisfying other kinds of color representations by using eitheruniformity(homogeneity) criterion which is based linear or nonlinear transformation.on features derived from spectral components.Thesecomponents are defined in a chosen color space 2.1 Linear Transformationsmodel. 2.1.1 YIQ It has long been recognized that human eye YIQ is used to encode color information in TVcan discern thousand of color shades and intensities signal for American system. It is obtainedbut only two dozenshades of gray. Compared to from the RGB model by a linear transformation.gray scale, color provides information in addition tointensity. Color is useful or even necessary for Y 0.299 0.587 0.114 Rpattern recognition and computer vision. I = 0.596 -0.274 -0.322 G Color image segmentation attracts more Q 0.211 -0.253 -0.312 Band attention mainly due to following reasons. 1) color Images can provide more information than gray level images. Where 0 ≤ R ≤ 1, 0 ≤ G ≤ 1, 0 ≤ B ≤ 1. 2) the power of personal computer is 2.1.2 YUV increasing rapidly, and PC’s can be used to process YUV is also a kind of TV color representation color images now. suitable for European TV system. The transformation is This paper is organized as follows: firstly different color spaces are discussed, then main Y 0.299 0.587 0.114 R image segmentation algorithms are classified, then U = -0.147 -0.289 0.437 G advantages and disadvantages of different image V 0.615 -0.515 -0.100` B 1277 | P a g e
  • 2. Sameena Banu / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.1277-1281 Where 0 ≤ R ≤ 1, 0 ≤ G ≤ 1, 0 ≤ B ≤ 1. There are a number of CIE spaces can be created2.1.3 I1I2I3 once the XYZ tristimulus coordinates are Known. CIE (L* a* b*) space and CIE (L* u* v*) space areI1= ( R+G+B ) / 3 , I2= ( R-B ) / 2, two typical examples. They can be obtained through I3= ( 2G-R-B ) / 4 nonlinear transformation of X,Y,Z values: i) CIE(L* a* b*) is defined as: Comparing I1I2I3 with 7 other color spaces L* = 16(Y/Y0)1/3 – 16(RGB, YIQ, HIS, Nrgb, CIE xyz, CIE(L*u*v*) and a* = 500 [ (X/X0) 1/3- (Y/Y0)1/3]CIE(L*a*b*) , claimed that I1I2I3 was more effective b* = 200 [ (Y/Y0)1/3 – (Z/Z0)1/3]in terms of the quality of segmentation and where fraction Y/Y0 > 0.01 , X/X0 > 0.01computational complexity of the transformation. and Z/Z0>0.01 (X0,Y0,Z0) are (X,Y,Z) values for standard2.2 Nonlinear transformations white.2.2.1 Normalized RGB (Nrgb) The normalized color space is formulated ii) CIE (L* u* v*) is defined as: as: L* = 116( Y/Y0)1/3-16 r = R / (R+G+B) g = G / (R+G+B) u* = 13 L* (u’- u0) b = B / (R+G+B) v* = 13 L* (v’- v0) Since r+g+b=1 whe=n= two components Where Y/Y0 >0.01, Y0,u0 and v0 are the values ofare given , the third component can be determined. standard white and u’ =4X/(X+15Y+3Z), v’=6Y(X+15Y+3Z)2.2.2 HSIHue represents the basic colors, and is determined 3. CLASSIFICATION OF COLORby dominant wavelength in the spectral distribution SEGMENATATION ALGORITHMSof light wavelength. The saturation is the measure of Image segmentation is a process ofthe purity of the color, and signifies the amount of dividing an image into different regions such thatwhite light mixed with the hue. I is the average each region is, but the union of any two adjacentintensities of R, G and B. regions is not, homogenous. A formal definition ofThe HIS coordinates can be transformed from the image segmentation is as follows[1]: If p( ) is aRGB space as: homogeneity predicate defined on groups of connected pixels, then segmentation is a partition ofH = arctan(sqrt(3)(G-B) / ( R-G ) + ( 2R – G – B)) the set F into connected subsets or regions ( S1,S = 1 – min(R,G,B) / I S2,……..,Sn) such thatI = ( R+G+B) / 3 Uni=1 Si = F , with Si ∩ Sj =Φ (i≠ j) The uniformity predicate P(Si) = true for all regions,2.2.3 HSV Si, and P(Si U Sj ) = false, when i≠ j and Si and SjThe HSV color space is similar to the HIS space are neighbors.while the value (V) component is given by an Most of segmentation algorithms are basedalternate transformation as the maximum of the on two characters of pixel gray level: discontinuityRGB component. around edges and similarity in the same region. There are three main categories in imageH = arctan(sqrt(3)(G-B) / ( R-G ) + ( 2R – G – B)) segmentation : A. Edge-based segmentation ; B.S = 1 – 3*min(R,G,B) / (R+G+B) Region-based segmentation; C. Special theory basedV = max( R,G,B) segmentation.2.2.4 CIE Space 4. EDGE-BASED SEGMENTATIONCIE (Commission International de 1’Eclairage) In a color image, an edge should be definedcolor system was developed to represent perceptual by discontinuity in a three dimensional color spaceuniformity and thus meet the psychophysical need and is found by defining a metric distance in somefor a human observer. It has three primaries denoted color space and using discontinuities in the distanceas X,Y and Z. Any color can be specified by to determine edges. Another way to find an edge incombination of X,Y,Z. The values of X,Y,ZZ can be a color image is by imposing some uniformitycomputed by a linear transformation from RGB constraints on the edges in the three colorcoordinates. components to utilize all of the color components simultaneously, but allow the edges in the three X 0.607 0.174 0.200 R color components to be largely independent. Y = 0.299 0.587 0.114 G For each color space, edge detection is Z 0.000 0.066 1.116 B performed using the gradient edge detection method. The color edge is determined by the maximum 1278 | P a g e
  • 3. Sameena Banu / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.1277-1281values of 24 gradient values in three componentsand eight directions at a pixel. There are three most A. Fuzzy Techniquescommonly used gradient based methods: differential Fuzzy set theory provides a mechanism tocoefficient technique , Laplacian of Gaussian and represents and manipulate uncertainty andCanny technique. Among them Canny technique is ambiguity. Fuzzy operators, properties,most representative one. mathematics, and inference rules have found considerable applications in image5.REGION-BASED SEGMENTATION segmentation.[3,4,5,6]. Prewitt first suggested that Edge-based segmentation partition an the output of image segmentation should be fuzzyimage based on abrupt changes in intensity near the subset rather than ordinary subsets. In fuzzy subsets,edges whereas region-based segmentation partition each pixel in an image has a degree to which itan image into regions that are similar in according to belongs to a region or a boundary, characterized bya set of predefined criteria. Thresholding, region membership value.growing, region splitting and region merging are the Recently, there has been an increasing usemain examples of techniques in this category. of fuzzy logic theory for color image segmentation [7 ,8]. For color images colors tend to form clusters A. Thresholding Methods in color space which can be regarded as a natural feature spcace. Thresholding techniques are imagesegmentations based on image-space regions. The B. Physics based techniquesfundamental principle of thresholding techniques is Physics based segmentation techniquesbased on the characteristics of the image[2]. It involve physical models to locate the objects’chooses proper thresholds T n to divide image pixels boundaries by eliminating the spurious edges ofinto several classes and separate the objects from shadow or highlights in a color image. Among thebackground. When there is only a single threshold physics models, “dichromatic reflection model” [9]T, any point (x,y) for which f(x,y)>T is called an and “approximate color-reflectance modelobject point; and a point (x,y) is called a background (ACRM)”point if f(x,y)<T. T is given by: [10] are the most common ones.T = M[x , y , p(x,y) , f(x,y)], based on this equationthresholding technique can be divided into global C. Neural Network techniqueand local, thresholding techniques. Neural network based segmentation is totally different from conventional segmentation1) Global thresholding: When T depends only Algorithms. In this algorithm, an image is firstlyon f(x,y) and the value of T solely relates to the mapped into a neural network where every neuroncharacter pixels, this thresholding technique is stands for a pixel. Then , we extract image edges bycalled global thresholding techniques. There are using dynamic equations to direct the state of everynumber of thresholding techniques such as: neuron towards minimum energy defined by neuralminimum thresholding, Otsu, optimal thresholding, network [11]. Neural network based segmentationhistogram concave analysis, iterativw thresholding, has three basic characteristics: 1) highly parallelentropy-based threshholding. ability and fast computing capability, which make it suitable for real-time application; 2) unrestricted and2) Local thresholdin: If threshold T depends nonlinear degree and high interaction amongon both f(x,y) and p(x,y), this thresholding is called processing units, which make this algorithm able tolocal thresholdind. Main local thresholdind establish modeling for any process; 3) satisfactorytechniques are simple statistical thresholding, 2-D robustness making it insensitive to noise. However,entropy-based thresholding, histogram there are some drawbacks for neural network basedtransformation thresholding etc. on segmentation, such as: 1) some kinds of segmentation information should be known6. SPECIAL–THEORY BASED beforehand; 2) initialization may influence the result SEGMENTATION of image segmentation; 3) neural network should be Numerous special-theory based trained using learning process beforehand, thesegmentation algorithms derive from other fields of period of training may be very long, and we shouldknowledge avoid overtraining at the same time.such as wavelet transformation, morphology, fuzzymathematics, genetic algorithm, artificialintelligence and so on. 1279 | P a g e
  • 4. Sameena Banu / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.1277-1281 7. Table. Monochrome image segmentation techniquesSegmentation Method Description Advantages DisadvantagesTechniqueHistogram Requires that the It does not need a prior 1)Does not work well for anThresholding histogram of an image information of the image. image without any obvious peaks has a number of peaks, And it has less or with broad and flat valleys; each correspond to a computational complexity 2)Does not consider the spatial region. details, so cannot guarantee that the segmented regions are contiguous.Region-Based Group Pixels into Work best when the region 1)Are by nature sequential andAppraoches homogeneous regions. homogeneity criterion is quite expensive both in Including region easy to define. They are computational time and memory; growing, region splitting, also more noise immune 2)Region growing has inherent region merging or their than edge detection dependence on the selection of combination. approach. seed region and the order in which pixels and regions are examined; 3)The resulting segments by region splitting appear too square due to the splitting scheme.Edge Detection Based on the detection of Edge detection technique 1)Does not work well with imagesApproach discontinuity, normally is the way in which human in which the edges are ill-defined tries to locate points with perceives objects and or there are too many edges; more or less abrupt works well for images 2)It is not a trivial job to produce a changes in gray level. having good contrast closed curve or boundary; between regions. 3)Less immune to noise than other techniques.Fuzzy Apply fuzzy operators, Fuzzy membership 1)The determination of fuzzyApproaches properties, mathematics, function can be used to membership is not a trivial job; and inference rules, represent the degree of 2)The computation involved in provide a way to handle some properties or fuzzy approaches could be the uncertainty inherent linguistic phrase, and intensive. in a variety of problems fuzzy IF-THAN rules can due to ambiguity rather be used to perform than randomness. approximate inference.Neural Network Using neural networks to No need to write 1)Training time is long;Approaches perform classification or complicated programs. 2)Initialization may effect the clustering Can fully utilize the result; parallel nature of neural 3)Overtraining should be avoided. networks. the results can be combined in some way to obtain final segmentation result. 8. CONCLUSSION In this paper, we classify and discuss colorspaces and main segmentation algorithms. Animage segmentation problem is basically one ofpsychophysical perception, and it is essential to References:supplement any mathematical solutions by a prior [1] H. D. Cheng, X.H.Jiang, Y.Sun and Jingknowledge about the picture knowledge. Most gray Li Wang, “Color Image Segmentation :level image segmentation techniques could be Advancesextended to color image, such as histogram & Prospects”.thresholding, region growing, edge detection and [2] Gao Xinbo, Fuzzy cluster Analusis andfuzzy based approaches. They can be directly application, Sian:Xidian Universityapplied to each component of a color space, then press,2004, pp. 124-130. 1280 | P a g e
  • 5. Sameena Banu / International Journal of Engineering Research and Applications (IJERA) ISSN: 2248-9622 www.ijera.com Vol. 2, Issue4, July-August 2012, pp.1277-1281[3] S.K. Pal et al., “A review on Image segmentation techniques’, Pattern Recognition, 29, 1277,1294, 1993.[4] J. M. Keller, and C. L. Carpenter, “Image Segmentation in the presence of Uncertainty” International Journal if Intelligent Systems, Vol. SMC-15, 193- 208, 1990.[5] S. K. Pal and R. A. King, “ On Edge Detection of X-ray Images Using Fuzzy Sets”, IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol. PAMI-5, No.1, 69-77, 1983.[6] I. Bloch, “Fuzzy Connectivity and Mathematical Morphology”, Pattern Recognition letters, Vol. 14, 483-488, 1993.[7] G.S. Robinson, “Colr Edge Detection”, Optical Engineering, Vol. 16, No. 5, Sept. 1977.[8] T. L. Huntsberger, C. L. Jacobs and R. L. Canon, “Iterative Fuzzy Image Segemntation”, Pattern Recognition, Vol. 18, No. 2, 131-138, 1985.[9] S. A. Shafer, “Using Color to separate Reflection Components”, Color Reseach and Application, 210-218, 1985.[10] G. Healey, “Using Color for Geometry- insensitive Segmentation”, Optical Society of America, Vol, 22, No. 1, 920- 937, 1989.[11] Wang Qiaoping, “One image segmentation technique based on wavelet analysis in the context of texture”, Data Collection and Processing, 1998,vol.13, pp. 1216.[12] Wladyslaw Skarbek and Andress Koschan, “Colour Image Segmentation- A Survey”. 1281 | P a g e

×