8 ijaers dec 2014 10 image segmentation techniques γçô a review

International Journal of Advanced Engineering Research and Science (IJAERS)

[Vol-1, Issue-7, Dec.- 2014] ISSN: 2349-6495

Image Segmentation Techniques – A Review Praveen Agrawal1, Ashok Kajla2, Manish.N.Raverkar3 1

M.Tech (pursuing), Dept. of Electronics & Communication, Arya Institute of Engineering & Technology, Kukas, Jaipur

2,3

Associate Professor, Dept. of Electronics & Communication, Arya Institute of Engineering & Technology,Kukas Jaipur,

India India

Abstract—Image segmentation can be a traditional theme in the field of image processing as well as is a hot spot while focusing involving image processing methods.

While using the improvement of personal computer processing abilities plus the improved application of color image, the actual color image segmentation are more and more implicated from the scientists. Color image segmentation techniques is visible as an expansion in the gray image segmentation method within the color graphics, many the main gray image segmentation methods is not immediately put on color photos. This involves to improve this method involving unique gray image segmentation method using the color image that are fitted with the feature involving plenteous information or perhaps search a whole new image segmentation method this especially utilised in color image segmentation. This paper presents a review on different image segmentation techniques and a comparative study on these techniques.

I. INTRODUCTION Images are a type of information which defined as a function of f(x,y) where x, y are spatial coordinates, and the amplitude of the function f(x, y) is called intensity or gray level of the image at the point [1]. Image segmentation is a technique and process which divide the image into different feature of region and extract out the interested target. Here features can be pixel gray scale, color, texture, etc. Pre-defined targets can correspond to a single region or multiple regions.[3] Image segmentation is a classic subject in the field of image processing and also is a hotspot and focus of image processing techniques. Color image segmentation methods can be seen as an extension of the gray image segmentation method in the color images, but many of the original gray image segmentation methods cannot be directly applied to color images [2]. In order to segment an image we need some mathematical representation of pixels and then develop an algorithm to compute the segments [3]. In next section we brief some approaches for image segmentation.

II.

APPROACHES OF IMAGE SEGMENTATION Most of segmentation algorithms are based on two characters of pixel gray level: discontinuity around edges and similarity in the same region. There are three main categories in image segmentation : A. Edge-based segmentation ; B. Region-based segmentation; C. Special theory based segmentation.

2.1 EDGE-BASED SEGMENTATION In a color image, an edge should be defined by discontinuity in a three dimensional color space and is found by defining a metric distance in some color space and using discontinuities in the distance to determine edges. Another way to find an edge in a color image is by imposing some uniformity constraints on the edges in the three color components to utilize all of the color components simultaneously, but allow the edges in the three color components to be largely independent. For each color space, edge detection is performed using the gradient edge detection method. The color edge is determined by the maximum values of 24 gradient values in three components and eight directions at a pixel. There are three most commonly used gradient based methods: differential coefficient technique Laplacian of Gaussian and Canny technique. Among them Canny technique is most representative one.[8] 2.2 REGION-BASED SEGMENTATION Edge-based segmentation partition an image based on abrupt changes in intensity near the edges whereas regionbased segmentation partition an image into regions that are similar in according to a set of predefined criteria. a region growing algorithm starts with a seed point or seed area and then progressively evaluates and adds or discards neighbours to the region based on their similarity to the region until a stopping criterion is met.[6] Thresholding, region growing, region splitting and region merging are the main examples of techniques in this category.

A. Thresholding Methods Page | 44

International Journal of Advanced Engineering Research and Science (IJAERS)

Thresholding techniques are image segmentations based on image-space regions. The fundamental principle of thresholding techniques is based on the characteristics of the image[2]. It chooses proper thresholds Tn to divide image pixels into several classes and separate the objects from background. When there is only a single threshold T, any point (x,y) for which f(x,y)>T is called an object point; and a point (x,y) is called a background point if f(x,y)<T. T is given by: T = M[x , y , p(x,y) , f(x,y)], based on this equation thresholding technique can be divided into global and local, thresholding techniques.[8] Some fuzzy techniques are applied for better result for image segmentation. Fuzzy set theory provides a mechanism to represents and manipulate uncertainty and ambiguity. Fuzzy operators, properties, mathematics, and inference rules have found considerable applications in image segmentation. Prewitt first suggested that the output of image segmentation should be fuzzy subset rather than ordinary subsets. In fuzzy subsets, each pixel in an image has a degree to which it belongs to a region or a boundary, characterized by membership value.

Recently, there has been an increasing use of fuzzy logic theory for color image segmentation. For color images colors tend to form clusters in color space which can be regarded as a natural feature space.

III.

[Vol-1, Issue-7, Dec.- 2014] ISSN: 2349-6495

on the above homogeneity/heterogeneity assumption for the image regions. The proposed noise reduction technique is applied locally by processing the neighborhood of each pixel separately. The underlying idea is estimating the true pixel intensity by detecting the presence (or absence) of image structure (homogeneity versus heterogeneity) and by applying the appropriate estimation technique. At the second stage, the gradient of the smoothed image is calculated using the Gaussian filter derivatives with a small scale since the noise has already been substantially reduced at the first stage. Then, the gradient magnitude is calculated and thresholded appropriately. At the next stage, the resulting gradient magnitude is passed on to the watershed detection algorithm, which produces an initial image partition. At the final stage, a novel fast region-merging process is applied to the most similar pair of adjacent regions at each step. The merging process may be terminated either interactively or with the use of a given stopping rule based on hypothesis testing. The proposed solution to accelerate region merging is based on the observation that it is not necessary to keep all RAG edges in the heap but only a small portion of them [7]

DIFFERENT ALGORITHMS FOR IMAGE SEGMENTATION

The paper[4] presents the series of algorithms on watersheds. The pre processing stage involved is an edgepreserving statistical noise reduction approach for the accurate estimation of the image gradient. Then, initial partitioning of the image into primitive regions is carried out by applying the watershed transform on the image gradient magnitude. . This initial segmentation is the input to a computationally efficient hierarchical (bottom up) region merging process to produce an output as the final segmentation. Further this process is accompanied by region adjacency graph (RAG) representation of those image regions produced. Then the most similar pair of regions at each step is chosen (minimum cost RAG edge), to be merged accompanied by updation of corresponding RAG. Traditional way of implementation included sorting all RAG edges in a priority queue. The author propose a significantly faster algorithm, which additionally maintains the so-called nearest .neighbour graph, due to which the priority queue size and processing time are drastically reduced. According to the propose algorithm The aim of the first stage is the reduction of the noise corrupting the image while preserving its structure, based

Fig 1 Flow chart diagram Fast Nearest Neighbor Region Merging Algorithm A another technique for Quaternion Framework for Color Image Smoothing and Segmentation the author using Hamiltonian quaternion framework. The key innovation for this work here is a unified approach to color image processing using (i) a novel quaternion Gabor filter (QGF) to extract the local orientation, and (ii) continuous mixtures on the unit sphere to model the derived orientation for developing spatially varying smoothing and segmentation kernels. In this the author First, introduce a filter named quaternionic Gabor filter (QGF) which can combine the color channels and the Page | 45

International Journal of Advanced Engineering Research and Science (IJAERS)

orientations in the image plane which are optimally localized both in the spatial and frequency domains and provide a good approximation to quaternionic quadrature filters. the local orientation information in the color images are extracted. Second, in order to model this derived orientation information, continuous mixtures of appropriate exponential basis functions and derive analytic expressions for these models. These analytic expressions take the form of spatially varying kernels which, when convolved with a color image or the signed distance function of an evolving contour (placed in the color image), yield a detail preserving smoothing and segmentation, respectively. In the paper the author first describe the quaternion algebra and quaternion Fourier transform, and then present a novel definition of the QGFs. After this author introduce a continuous mixture model for quantifying the derived orientation information to perform color image smoothing. Next author propose another continuous mixture model on the derived orientation information for use in segmentation. The proposed methods handle the coupling between the channels through the application of QGFs to the quaternion representation of color images. This process also integrates the color information and the texture information. In segmentation, the active contour evolution is controlled by this unified information. Similarly, in the denoising task, image channels do not evolve independently because the orientation space and the color components are linked through the QGFs.[5] IV. CONCLUSION In this paper we go through with some fundamental Edge Based segmntation or Region based segmentation methods for image segmentation like and then we study two different methods of image segmentation. The Quaternion Framework method handle coupling between the channels through filter and also integrates the color information and the texture information. The Fast Nearest Neighbor Region Merging approach allows the user to select the iteration at which the resulting segmentation is acceptable. But some demerits are that the memory requirements are relatively high due to the watershed detection algorithm[9] and the noise reduction stage is not so efficient when SNR is low.

[Vol-1, Issue-7, Dec.- 2014] ISSN: 2349-6495

[2] Jun Tang A Color Image Segmentation algorithm Based on Region Growing 978-1-4244-6349-7 in 2010 [3] S. Wesolkowski, M.E. Jernigan, R.D. Dony, "Global Color Image Segmentation Strategies: Euclidean Distance vs. Vector Angle," in Y.-H. Hu, J. Larsen, E. Wilson and S. Douglas (eds.), Neural Networks for Signal Processing IX, IEEE Press, Piscataway, NJ, 1999, pp. 419-428.

[4] Kostas Haris, Serafim N. Efstratiadis, Nicos Maglaveras, and Aggelos K. Katsaggelos, “Hybrid Image Segmentation UsingWatersheds and Fast Region Merging”in IEEE Transactions on image processing, vol. 7, no. 12, december 1998.

[5] Özlem N. Subakan · Baba C. Vemuri “A Quaternion Framework for Color Image Smoothing and Segmentation” Int J Comput Vis (2011) 91: 233– 250DOI 10.1007/s11263-010-0388-9 [6] Mariusz Nieniewski Extraction of Diffuse Objects From Images by Means of Watershed and Region Merging: Example of Solar Images in IEEE Transactions On Systems, Man, And Cybernetics— Part B: Cybernetics, Vol. 34, No. 1, February 2001 [7] K. Haris, “A hybrid algorithm for the segmentation of 2D and 3D images,” Master’s thesis, Univ. Crete, 1994.

[8] Sameena Banu “ The Comparative Study on Color Image Segmentation Algorithms “in International Journal of Engineering Research and Applications ISSN: 2248-9622 Vol. 2, Issue4, July-August 2012, pp.1277-1281 [9] L. Vincent and P. Soille, “Watersheds in digital spaces: An efficient algorithm based on immersion simulations,” IEEE Trans. Pattern Anal. Machine Intell., vol. 13, pp. 583–598, June 1991.

REFERENCES [1] R. C. Gonzalez and R. E. Woods, Digital Image Processing, Pearson Education 2nd Edition Asia, 2002. ISBN: 0-201-18075-8

Page | 46