Rank based Approach for Extracting Unit Pixel Width Skeleton

Skeletonization is a transformation of a digital picture component into a subset of the original component. The skeleton is a qualitative tool for matching and analyzing 2D objects. Skeletonization is considered as an important issue because of the close link between an object’s skeleton and its boundary. Arank based skeleton extraction algorithm is presented which iterates itself to remove all the skeleton branches corresponding to the unimportant shape regions depending on the boundary extraction method. The method proposed performance is compared with various methods in the literature such as ZS, LW, and MZS skeletonization algorithms and identified that proposed method out performs well with respect to the performance measures elapsed time, thinning rate, thinning speed, and connectivity.


Introduction
Skeletonization, the extract of object skeletons from a given digital image. It is morphological operation that iteratively removes black foreground pixels by layer until 1-point width is achieved. We can redefine the skeleton extraction or reduce a digital image to a minimum size or to reduce the image to such a degree that an image preserves the points required for image transformation. The performance can be assessed using the following parameters [7]: -Elapsed Time, Thinning Speed, Thinning Rate, Connectivity and the image should not contain noise and unnecessary branches.
Altogether, even though more than 300 algorithms were proposed for skeletonization, improvement is required because the current approximate skeletonization Algorithms frequently present one or more of the following drawbacks: x A high-resolution picture can take a long time to skeletonize.
x Skeletons may not be based inside the form below.
x The obtained skeletons are susceptible to changes in the noise and structure, such as rotation or scaling. x There may be a different number of components in a shape and its skeleton.
x Objects such as noisy spurts and false short branch between the break can be contained in skeleton x Points of intersection.
x The branches of the skeleton may be weakened greatly. Moreover, most techniques are acceptable not for grey images but only for binary images.
Many approach methods for the measurement of skeletons in the discrete world were proposed which can be divided in two groups: the pixel-based and the pixel-based process. Thinning and distance transformation are techniques used in pixel-based approach [11]. There are two types of non-pixel approaches, based either on cross-section and diagrams of Voronoi. [15].
Thinning is a major preprocessing technique important for applications such as signature checks, pattern recognition and compression, etc. Thinning algorithm is a morphological procedure that extracts from binary images selected pre-figure pixels. It retains the topology (scope and connectivity) of the original region while removing most of the original pixels. Thinning makes it easy and convenient to remove and identify functionality by reducing a certain pattern to a unit thickness. The diluted process converts the figure1 into a digital pattern with lower thicknesses from one type to another. The rest of the paper is structured as: presentation of the motivation with the existing methods in section 2. The proposed method of image retrieval is modeled in section 3, which insists the significance of the method for skeletonization. The results of the method and performance analysis is demonstrated in section4 and finally, section 5 concludes the paper.

Related Work
There are three major techniques for skeletonization: x Distance map of ridges of border points identify, x calculating the boundary point diagram of Voronoi, and x Layer erosion known as thinning The lower point of a binary image is called black in Distance Transform [16] with a pixel value of 0; the object value is white and the pixels are 1. A binary image is to be converted to another image, which offers a value that corresponds to a minimum distance of the background for each object pixel. Similar distance metrics lead to various transformations in space. Point p is a pixel target and q is the nearest pixel background. The coordinates of p and q are (x p ,y p) and (x q ,y q) . The Euclidean distance is defined as: The drawback of distance transformation is that the extracted skeleton does not allow for connectivity and completeness (i.e. the extracted branches can be separated, not all significant visual sections can be represented). Thinning algorithms [1] [3] can be split into iterative algorithms and not iterative algorithms. Noniterative algorithms are fast, but they produce inexact results. Reliable and efficient, iterative algorithms are thinner. The thinning procedure uses templates, where the centre pixel is removed by a match between the template in the image. Iterative [4][4] [6] algorithms erode pixel outer layers until layers are no longer available. Nearly all iterative thinning algorithm models, including the Stentiford Thinning process, use Identifying and extracting models [16].
A very successful and proven fine algorithm, Zhang and Suen [1] is the ZSA algorithm [17] that has been proposed by both Zhang and Suen. The algorithm runs concurrently on three neighborhoods. It is thinning. The ZS algorithm has two subphrases and is a directional algorithm. The first subitem tries to delete pixels in the southwestern frontier and pixels in the northwestern corner, while the other is intended to remove pixels in the northwest borders and in the southeast corner (i.e. opposite lines). Connectivity failure due to the absolute absence of 2X2 quadrant patterns is the downside of this algorithm. This does not cover the topology. The results are not a pixel-thick skeleton that breaks up the first property of a successful thinning algorithm. The redundant pixels are responsible. In the LW algorithm (Lu and Wang) [14], U and Wang proposed an enhanced ZS algorithm, known as LW, to resolve the issue of excessive tilt line erosion by changing the neighborhood state by changing the region. This algorithm is also influenced by the problem of the 2X2 square

Rank based Skeleton
The following are the steps to extract the skeleton from original image.

Generate initial skeleton
Input: original image output: S(A) A skeleton S(A) is simple. If z is a point of S(A) and (D)z is the largest disk centered in z and contained in A -this disk is called "maximum disk". The disk (D)z touches A's boundary at two or more different locations. The skeleton of A is defined by terms of erosions and openings k times, and K is the last iterative step before A erodes to an empty set, in other words: In conclusion S(A) can be obtained as the union of skeleton subsets S k (A). The skeleton can be obtained using the operations of mathematical morphology.

Generate initial polygon
Input: A output: P n initial polygon The boundary of an image is regarded as the initial polygon P n A convex hull is used for identifying the boundary of an image. Using morphological algorithms Convex hull is generated. The convex hull of set A is generated with the use of structuring element sequences. Let Bi, i=1 2, 3, 4 stand for 4 structuring elements as shown in figure 2. The method involves applying the hit-or -miss transform iteratively to A with B1, but if further adjustments do not occur the union takes place with A and the outcome is called D1. The method is repeated with B2 until there are no additional changes. The union of four resulting Ds is the convex hull of A.

Generate simple polygon
Input: P n output:P k A consecutive segment of the convex hull is called i.e. S1, S2 is substituted with an individual line section, which links the S1 U S2 endpoints. DCE generates a basic polygons series P = Pn, Pn-1, ---,

Performance Analysis Measures
Performance assessment is an important criterion for the implementation of any algorithm in a given application. Many smaller algorithms and techniques available require trade-offs between one or more topological and geometrical demands. Thinning algorithms can be estimatedaccording to the following parameters:  For each algorithm applied in the thinning process, elapsed time, thinning rate, thinning speed, and connectivity are measured. Table Ishows obtained performance metrics by applying existing methods ZS, LW, MZS and proposed Rank based algorithm. Here we obtained elapsed time, thinning rate, thinning speed and connectivity to measure the performance of proposed method in comparison with existing methods. The performance metrics are calculated by applying ZS, LW, MZS, proposed Rank based on around 1400 images containing 70 classes of shapes by taking 20 image members from each class. In Table  I, obtained measures are shown only for 10 image classes like bird, carriage, butterfly, camel, beetle, bottle, brick, children, face, and hammer. It is observed that the proposed method exhibits better performance with elapsed time with most of the cases. Thinning rate is remained same with proposed method in comparison with existing methods. The proposed method exhibits improved thinning speed with image classes belonging to bird, carriage, butterfly, camel, beetle, bottle, brick, children, but only average thinning speed is obtained with image classes face, hammer in comparison with existing methods. Even though, thinning speed is good with face and hammer image classes in existing methods but those methods are leaded to higher value in elapsed time.
Improved connectivity measure is achieved through proposed rank based skeleton algorithm with image classes butterfly, camel, beetle containing complex shape structures, for these images, the ICRAEM 2020 IOP Conf. Series: Materials Science and Engineering 981 (2020) 032005 IOP Publishing doi:10.1088/1757-899X/981/3/032005 9 existing methods yielded connectivity value in terms of hundreds but proposed method has exhibited a value not more than 10 even with complex classes.
Comparing with all 70 classes of images with complex structures, we have observed the proposed method exhibits better performance measures in comparison with ZS, LW, and MZS thinning algorithms.

Conclusion
In this paper, rank based skeletonization technique is presented. The skeletons are obtained using the boundary based on polygon simplification method. The experiments are conducted on MPEG7 datasets which shows that proposed rank based skeletonization method is significantly superior to skeletonization methods listed in the literature. The performance is measured using both subjective and objective analysis. The Subjective analysis shows that skeleton has singleton potential. The objective analysis has been done against Thinning rate, Thinning speed, Connectivity and Elapsed time. It was observed that Rank Based Skeletonization technique gives good results for all parameters above other algorithms.