Abstract
This paper introduces a modified method for construction of a new design-control 6-point approximating refinement scheme. The construction of the new scheme is based on translation of the points of 4-point approximating refinement scheme to the new position according to the linear combination of certain displacement vectors. The initial and terminal points of these vectors are the refinement points of two 4-point approximating binary schemes. The new scheme contains three design-control parameters. These parameters increase the efficiency and flexibility of the new scheme. The mathematical and graphical analysis of the refinement scheme show that this scheme is good choice for curve modeling.
1. Introduction
Computer aided geometric design (CAGD) is considered as an emerging research field of computational mathematics hat has been rapidly growing since the past two decades due to a wide range of applications (Garg, Citation2020). In CAGD different mathematical algorithms and refinement rules are used to obtain better shapes of the objects. The CAGD field works on the same principle that was used by Archimedes to approximate a circle by increasing the number of vertices of a regular polygon. This field concerns with modeling and designing different complex objects with the help of elegant mathematical algorithms. Therefore, CAGD is the field which is mostly used by designers for shaping or fitting curves and surfaces with non-uniform shapes or discrete set of control points.
Refinement schemes have been a well-liked techniques in CAGD to produce curves and surfaces. Refinement schemes have various applications in graphics, image processing, engineering etc and thus have become an important area of study (Livesu, Citation2021; Livesu, Pitzalis, & Cherchi, Citation2021; Meng et al., Citation2020; Ning et al., Citation2020; Zhang et al., Citation2020). These schemes succeeded to get a great attention due to their high efficiency and clarity. A subdivision process refines the initial polygon recursively to a set of refined polygons which converge to a smooth limiting curve. Also, the refinement schemes were designed mainly for getting the desirable shapes. In addition, by different refinement schemes we can create smooth curves and surfaces by using a sequence of successive refinements. It starts by choosing a rough shape, and then converges finally to a smooth shape (Mahalingam & Koneru, Citation2020).
The most important classes of refinement schemes are interpolatory and approximating refinement schemes which interpolate and approximate the given data respectively. Both are considered as a non-parametric binary approximating refinement schemes. The construction and analysis of the non-parametric binary approximating refinement schemes are discussed in (Hameed & Mustafa, Citation2017; Mustafa & Rehman, Citation2010; Siddiqi & Younis, Citation2013). Also, the parametric binary approximating schemes and their inspections are presented in (Asghar & Mustafa, Citation2019; Daniel & Shunmugaraj, Citation2010; Rehan & Siddiqi, Citation2015; Siddiqi & Rehan, Citation2010; Citation2011).
In this paper, we present a novel method which merges two non-parametric binary refinement schemes to get a new parametric binary approximating refinement schemes called the design-control binary approximating refinement scheme. The schemes which are used in the construction process have equal complexity, but have different geometric behavior. Hence, the new scheme gives the geometric flexibility. Moreover, it contains the geometric behavior of both parent schemes as those are the special cases of the new design-control scheme.
The remainder of this research is organized as follows. In Section 2, we present some basic notations and results. In Section 3, the framework for the construction of the design-control binary refinement scheme is presented. In Section 4, we study the local control property of the design-control refinement scheme. We discuss the smoothness analysis of new refinement scheme in Section 5. And then, we give the Gibbs phenomena analysis of our refinement scheme in Section 6. Finally, summary and conclusions are given in Section 7.
2. Basic notation and results
A uni-variate linear binary refinement scheme is based on repeated (successive) application of the refinement rules. Which are used to map a polygon to a refined polygon The general compact form of these refinement rules is defined as: (2.1) (2.1) where denotes the space of scaler-valued sequences. The sequence is called the refinement mask. Therefore, the polynomial which uses this mask as coefficients is called the Laurent polynomial. In addition, the Laurent polynomial corresponding to refinement scheme Equation(2.1)(2.1) (2.1) is:
The necessary condition for the convergence of a binary refinement scheme are: (2.2) (2.2)
Which is equivalent to the following relations; (2.3) (2.3)
Definition 2.1.
A linear refinement scheme is called approximating refinement scheme if it can be written as:
This type of refinement schemes generates the limiting curves in which the control points of k-th polygon level are not included in the points of the -th polygon level.
Definition 2.2.
Convergence analysis by Eigenvalues method
We seek the complete demonstration about the necessary and sufficient conditions for uniform convergence of the refinement scheme. This analysis was firstly introduced by Doo and Sabin (Doo & Sabin, Citation1978), and then developed through the years (see for example (Xumin, Xiaojun, Xianpeng, & Cailing, Citation2013; Thorne, Citation2021)). As a matter of fact, the matrix formalism is the technique for derive the necessary conditions for a scheme to be it based mainly on the eigenvalues of the subdivision matrix.
Let the eigenvalues of the subdivision matrix be where and And we have the necessary conditions for the following properties:
1: shows kink, i.e. not
2: shows unbounded curvature.
3: shows mildly diverging curvature.
4: shows curvature bounded.
5: shows curvature to zero.
Theorem 2.1.
Let the matrix formalism to derive the necessary conditions for a scheme to be based on the eigenvalues of the subdivision matrix. If the limiting curve is continuity, the eigenvalues satisfy (Hassan, Ivrissimitzis, Dodgson, & Sabin, Citation2002): and
And a scheme will be uniformly convergent if and only if there is an integer such that
Theorem 2.2.
Let the refinement scheme with mask where satisfying the condition (Dyn & Levin, Citation2002) (2.4) (2.4)
If there exist an integer , such that , then the refinement scheme is -continuous, where and Laurent polynomial , so, . When
Let with is contractive. Then is also convergent with -continuous.
Theorem 2.3.
Given , let g be a function defined by (Amat, Ruiz, & Trillo, Citation2018) with and . Let be a uni-variate stationary refinement scheme with condition: where define as: and . Then, if ; and if h is sufficiently small, we have the following two conditions:
P 1. If , then with
P 2. If , there exists such that:
3. Construction structure of the new 6-point design-control refinement scheme
In this section, we describe the construction process of the new design-control binary approximating refinement scheme. The new scheme contains three design-control parameters which control the behavior of the limit curves. Therefore, the step by step procedure of the construction is given below:
3.1. Step 1
In this step, we take two already published binary approximating refinement schemes.
Firstly, the 4-point approximating refinement scheme which was presented by (Deslauriers & Dubuc, Citation1989; Mustafa & Rehman, Citation2010) is (3.1) (3.1) where are the refined points of the subdivided polygon by scheme Equation(3.1)(3.1) (3.1) if are the given points of the given polygon.
Secondly, the 4-point approximating B-spline refinement scheme which is constructed by B-spline basis function of degree-6 is (3.2) (3.2) where are the refined points of the subdivided polygon by scheme Equation(3.2)(3.2) (3.2) if are the given points of the given polygon.
3.2. Step 2
In this step, firstly we define two displacement vectors. The first displacement vector is denoted by and is shown in . We get this vector by subtracting refinement point defined in Equation(3.2)(3.2) (3.2) from the refinement point defined in Equation(3.1)(3.1) (3.1) . That is
Similarly, we get the second displacement vector which is shown in by subtracting refitment point defined in Equation(3.2)(3.2) (3.2) from the refinement point defined in Equation(3.1)(3.1) (3.1)
Hence the displacement vectors can be expressed in the linear combination of the given control points That is (3.3) (3.3)
If we increase/decrease the indices of refinement equations defined in Equation(3.1)–(3.2) and define four more displacement vectors in same manner, we get (3.4) (3.4)
3.3. Step 3
In this step, we use the properties of “vector addition” and “scalar multiplication” to get the resultant vectors. We use the six vectors defined in Equation(3.3)–(3.4) along with three scalars and Hence we get two resultant vectors denote by: and and they defined below: (3.5) (3.5)
3.4. Step 4
In this step, we calculate refinement rules of the new 6-point approximating refinement scheme by translating the refinement points of scheme Equation(3.1)(3.1) (3.1) . For the translation of these points we use the resultant vectors and which are defined in Equation(3.5)(3.5) (3.5) . Hence the first refinement rule is:
Similarly, the second refinement rule is:
Thus, the new design-control approximating scheme is: (3.6) (3.6) where,
Here, and are points at -th refinement level, whereas and are the scalars used as design-control parameters. Thus the sequence is the mask of the scheme Equation(3.6)(3.6) (3.6) . The new 6-point approximating refinement scheme is denoted by
If we put in the pair of refinement Equationequation (3.6)(3.6) (3.6) , then we get a system of refinement equations. Which can be written in a matrix form as:
Then, the mask of refinement scheme defined in Equation(3.6)(3.6) (3.6) is
Remark 3.1.
In this article, we use the “sub-scheme” word for the refinement scheme which is the special case of our refinement scheme defined in Equation(3.6)(3.6) (3.6) .
3.5. Sub-scheme of the refinement scheme
In this section, we present the sub-scheme, with design-control parameter of scheme that is defined in Equation(3.6)(3.6) (3.6) . We set in Equation(3.6)(3.6) (3.6) to get the following sub-scheme of scheme (3.7) (3.7) where
Where, the mask of above scheme is Hence, the mask of corresponding refinement scheme is,
Remark 3.2.
For the scheme the vectors defined in Equation(3.5)(3.5) (3.5) reduce to
3.6. Sub-scheme of the refinement scheme
In this section, we give the sub-scheme with design-control parameter of scheme defined in Equation(3.6)(3.6) (3.6) . We set and in Equation(3.6)(3.6) (3.6) then, to get the following scheme denotes by (3.8) (3.8) where, with is the mask of this refinement scheme. Thus, the mask of corresponding refinement scheme is,
Remark 3.3.
For the scheme the vectors which are defined in Equation(3.5)(3.5) (3.5) take the following form:
3.7. Refinement rules for boundary points when the given polygon is open
Here, we discuss the refinement rules for the open polygon of our new approximating refinement scheme Whenever, we have to smooth an open polygon, we use a specific relations which replace the unknowns point with known points to get the refinement rules for boundary points of our refinement scheme.
Let the indices of the given points be as shown in . Hence, the point and are unknowns. So, first we use the following relation to find out the unknown points (3.9) (3.9)
Then, we use the following relation to replace the unknown points with the known points And then, (3.10) (3.10)
Hence, the boundary refinement points for the open polygons are:
4. Support of design-control refinement scheme
In this section, we analyze the support of our design-control refinement scheme. The support of our refinement scheme shows that how better it locally controls the limiting curves. The support of our refinement scheme is small, hence our scheme has better local control on shapes. We calculate the support of our refinement scheme and its sub-schemes and which are defined in Equation(3.6)(3.6) (3.6) , Equation(3.7)(3.7) (3.7) and Equation(3.8)(3.8) (3.8) respectively. The support of these schemes is analyzed theoretically and graphically as follows.
Lemma 4.1.
If we use refinement scheme on initial data (4.1) (4.1)
Then, after first subdivision step the non-zero points are
Proof.
If we use and in Equation(3.6)(3.6) (3.6) , we get and
If we put and in Equation(3.6)(3.6) (3.6) , we get and
If we substitute Equation(3.6)(3.6) (3.6) by and we get and
For and in Equation(3.6)(3.6) (3.6) , we get and
If we put and in Equation(3.6)(3.6) (3.6) , we get and
If we put and in Equation(3.6)(3.6) (3.6) , we get and
If we use and in Equation(3.6)(3.6) (3.6) , we get and
If we use and in Equation(3.6)(3.6) (3.6) , we get and
Hence, if we use the design-control refinement scheme defined in Equation(3.6)(3.6) (3.6) on the given initial data Equation(4.1)(4.1) (4.1) , then after first subdivision step the non-zero points are …, and □
Lemma 4.2.
If we use refinement scheme on initial data
Then, after second subdivision step, the non-zero points are
Lemma 4.3.
If we use refinement scheme on initial data
Then, after third subdivision step, the non-zero points are
Theorem 4.4.
The support width of our binary refinement scheme defined in Equation(3.6)(3.6) (3.6) is 11. So, it vanishes outside the closed interval [−5.5,5.5].
Proof.
In order to prove the above result, we use the results of Lemmas 4.1, 4.2 and 4.3. Let us define a set such that
By Lemma 4.1, we get that: if we use one subdivision step on initial data by our design-control refinement scheme, the leftmost non-zero point is and the rightmost non-zero point is
By Lemma 4.2, we get that: if we use two subdivision steps on initial data by refinement scheme the leftmost non-zero point is and the rightmost non-zero point is
Similarly, from Lemma 4.3, we get that: if we use three subdivision steps on initial data by scheme the leftmost non-zero point is and the rightmost non-zero point is
If we continue this process, then after k subdivision steps, the leftmost non-zero point is and the rightmost non-zero point is
The difference between left and right non-zero points at k-th subdivision steps is
If we get the support width of refinement scheme
Hence, the support width of our refinement scheme is 11. So, it vanishes outside the closed interval [−5.5, 5.5]. □
Corollary 4.5.
The support width of our refinement scheme is 7. So, it vanishes outside the closed interval [−3.5, 3.5].
Proof.
The proof of this corollary is trivial, because is a sub-scheme of scheme and it is obtained by putting in Equation(3.6)(3.6) (3.6) . □
Corollary 4.6.
The support width of our refinement scheme is 11. So, it vanishes outside the closed interval [−5.5, 5.5].
Proof.
The proof of this corollary is also trivial, because is a sub-scheme of scheme and it is obtained by putting and in Equation(3.6)(3.6) (3.6) . □
Now, we analyze the support of our refinement scheme graphically.
Experiment 4.1.
In this experiment, we plot an initial sketch by using data points . The curves generated by our refinement scheme are shown in . show the curves fitted by our refinement scheme after first, second and third refinement steps respectively. Here, the values of parameters are
Experiment 4.2.
In this experiment, we draw initial sketch by using the following initial control points . The black dashed-dotted lines in represent the initial sketch. While, red curve in is the curve fitted by our scheme after three refinement steps. Now we draw another initial sketch by changing the position of one point of the sketch which is shown in . The new sketch is shown in by black dashed-dotted lines. Which is simply obtained by moving point of previous sketch to the new position . shows the curve fitted by our scheme after three refinement steps. Here, the values of parameters are: . By comparing , we conclude that by moving one control point of the initial sketch, only a very small portion of limit curve is affected. This property shows that our refinement scheme has a good local control.
5. Continuity analysis of new design control refinement scheme
In this section, we present the convergence and smoothness analysis of the sub-schemes and of new design-control refinement scheme. For the analysis, we use the theoretical results of (Hassan et al., Citation2002) and (Dyn & Levin, Citation2002) which are given in Theorem 2.1 and Theorem 2.2 respectively.
5.1. Continuity analysis of refinement scheme
Here, we analyzed the convergence of binary approximating refinement scheme by using Eigenvalues of the subdivision matrix. We also check the level of smoothness of our proposed scheme by using the Laurent polynomial method for and
Theorem 5.1.
The refinement scheme Equation(3.7)(3.7) (3.7) satisfies the necessary conditions for continuity if the range of the tension parameter is
Proof.
To find the convergence of the refinement scheme which is defined in Equation(3.7)(3.7) (3.7) , consider the local subdivision matrix T of the scheme If we put in the pair of refinement Equationequations (3.7)(3.7) (3.7) , then we get a system of refinement equations. Which can be written in matrix form as: where, and the eigenvalues of the subdivision matrix T are:
From Theorem 2.1, the necessary conditions for -continuity are: (5.1) (5.1)
Now, by putting in Equation(5.5)(5.5) (5.5) , and compare the values
After simplification we get:
Thus, we have
Which, implies that the given scheme is convergent in this interval □
Theorem 5.2.
The binary refinement scheme Equation(3.7)(3.7) (3.7) satisfies the sufficient conditions of -continuity for
Proof.
The Laurent polynomial of refinement scheme Equation(3.7)(3.7) (3.7) is (5.2) (5.2)
By Theorem 2.2. (5.3) (5.3) where, and, (5.4) (5.4)
Now, by putting and in Equation(5.3)(5.3) (5.3) , we get:
This implies that
The above relation is true if lies between 0.7142857143 to 1.857142857. Hence the scheme is -continuous. If we put and in Equation(5.3)(5.3) (5.3) , then, we obtain
This implies that
Hence, this scheme is -continuous for
Now we substitute and in Equation(5.3)(5.3) (5.3) , we get
This implies that
Thus the scheme is -continuous for □
In , we present the continuity results of our refinement scheme by Laurent polynomial method, that we have obtained by using results of Theorem 2.2.
5.2. Continuity analysis of the refinement scheme
In this section, we analyzed the convergence of binary approximating refinement scheme by analyzing the eigenvalues of the subdivision matrix. We also check the level of smoothness of our proposed scheme by using the Laurent polynomial method for and
Theorem 5.3.
The refinement scheme Equation(3.8)(3.8) (3.8) satisfies the necessary conditions for continuity if the range of the tension parameter is
Proof.
To find the convergence of the refinement scheme defined in Equation(3.8)(3.8) (3.8) , consider the local subdivision matrix of the scheme If we put in the pair of refinement Equationequations (3.8)(3.8) (3.8) , then we get a system of refinement equations. Which can be written in matrix form as: where and,
The eigenvalue of the above subdivision matrix are:
From Theorem 2.1, the necessary conditions for -continuity are: (5.5) (5.5)
Let
This implies that
Thus, we have (5.6) (5.6)
Let,
This, implies that
By combining the above we get result as: (5.7) (5.7)
Let,
This, implies that
By combining above we get the result (5.8) (5.8)
If we take the common part of all inequalities Equation(5.6)(5.8) (5.8) , Equation(5.7)(5.7) (5.7) and Equation(5.8)(5.8) (5.8) , we get the following common interval (5.9) (5.9)
Theorem 5.4.
The refinement scheme Equation(3.8)(3.8) (3.8) satisfies the sufficient conditions for -smoothness for
In , we summarize the continuity results of our refinement scheme by Laurent polynomial method which we have obtained by using results of Theorem 2.2.
Now, we give an experiment to show the smoothness property of our design-control refinement scheme geometrically.
Experiment 5.1.
In this experiment, we draw the initial control polygon by using there initial control points: and . We show the curve generated by our refinement scheme in . shows the initial control polygon. show the curves fitted by our refinement scheme after one, two and three refinement steps respectively. Where =
6. Gibbs phenomenon analysis
In this section, we theoretically and graphically analyze the Gibbs phenomenon of sub-schemes and By using the mask of these refinement schemes, we obtain certain conditions to analyze whether there exist a Gibbs phenomenon close to the discontinuous points or not ().
Theorem 6.1.
The binary approximating refinement scheme defined in Equation(3.7)(3.7) (3.7) does not produce Gibbs oscillation in the limiting curves for
Proof.
By Theorem 2.3, a stationary scheme is free from Gibbs oscillation near to the discontinuity if (6.1) (6.1) where, (6.2) (6.2) and,
Since, the Laurent polynomial of our scheme is: (6.3) (6.3)
Therefore, if we fix and put in Equation(6.2)(6.2) (6.2) , we get (6.4) (6.4)
Now, we discuss two different cases corresponding to two values of namely, and
Case 1:
The value of in this case is fixed, that is Therefore we have (6.5) (6.5)
If we use in Equation(6.5)(6.5) (6.5) , we get (6.6) (6.6)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This implies that
Thus, we have (6.7) (6.7)
If we put in Equation(6.5)(6.5) (6.5) , we get (6.8) (6.8)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This implies that
Thus, we have (6.9) (6.9)
If we put in Equation(6.5)(6.5) (6.5) , we get (6.10) (6.10)
If we substitute in Equation(6.5)(6.5) (6.5) , we get (6.11) (6.11)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This implies that
Thus, we have (6.12) (6.12)
By Combining Equation(6.7)(6.12) (6.12) , Equation(6.9)(6.9) (6.9) and Equation(6.12)(6.12) (6.12) , we get the following result
Case 2:
The value of in this case is fixed, that is Therefore we have (6.13) (6.13)
If we substitute in Equation(6.13)(6.13) (6.13) , we get (6.14) (6.14)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This, implies that
Thus, we have (6.15) (6.15)
Now, if we put in Equation(6.13)(6.13) (6.13) , we get (6.16) (6.16)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This implies that
Thus, we have (6.17) (6.17)
Now, we put in Equation(6.13)(6.13) (6.13) , we get
Now, we put in Equation(6.13)(6.13) (6.13) , we get (6.18) (6.18)
From Equation(6.1)(6.1) (6.1) , the following relation is required
This implies that
Thus, we have (6.19) (6.19)
Now, we combine the interval for from Equation(6.15)(6.19) (6.19) , Equation(6.17)(6.17) (6.17) and Equation(6.19)(6.19) (6.19) , we get the following result (6.20) (6.20)
By Combining (6) and Equation(6.20)(6.20) (6.20) , we get the common interval for which our scheme does not produces Gibbs oscillations. □
Theorem 6.2.
The binary approximating refinement scheme defined in Equation(3.8)(3.8) (3.8) does not produce Gibbs oscillation in limiting shapes for
Proof.
We can prove this theorem by adopting the steps of Theorem 6.1. □
Experiment 6.1.
Here, in this experiment, we take the initial data from discontinuous function defined as: (6.21) (6.21)
Function Equation(6.21)(6.21) (6.21) provides us the initial data which is given in the and shown in . show the Gibbs oscillation free curves fitted by the scheme for and respectively.
Experiment 6.2.
In this experiment, we plot an initial sketch by using the data points and . We present the curves generated by our refinement scheme in . show the smooth and oscillation free curves fitted by our refinement scheme when we set and respectively.
7. Summary and conclusions
The summary of this research is:
We construct a binary 6-point approximating refinement scheme by using two existing binary 4-point approximating refinement schemes.
The support width of our scheme is 11 and its support region is
Our refinement scheme produces limiting curves up to smoothness.
If we set the value of parameters in the proved fixed range, our scheme produces curve free from artifacts and Gibbs oscillations.
Hence, we conclude that our scheme is a good choice for curve modeling with small support and high continuity, while the extra benefits can be achieved by using special values of parameters.
Disclosure statement
The authors declare that there are no conflicts of interest regarding the publication of this paper.
Data availability
The data used to support the findings of the study is available within this paper.
References
- Amat, S., Ruiz, J., & Trillo, C. (2018). Analysis of the Gibbs phenomenon in stationary subdivision scheme. Applied Mathematics Letters, 76, 157–163. doi:10.1016/j.aml.2017.08.014
- Asghar, M., & Mustafa, G. (2019). A family of binary approximating subdivision schemes based on binomial distribution. Mehran University Research Journal of Engineering & Technology, 38(4), 1087–1100. doi:10.22581/muet1982.1904.20
- Daniel, S., & Shunmugaraj, P. (2010). An extended three point approximating subdivision scheme, 2010 International Conference On Computer Design and Applications, Qinhuangdao, China, 2, 73–77.
- Deslauriers, G., & Dubuc, S. (1989). Symmetric iterative interpolation processes. Constructive Approximation, 5(1), 49–68. doi:10.1007/BF01889598
- Doo, D., & Sabin, M. A. (1978). Behaviour of recursive division surfaces near extraordinary points. Computer Aided Geometric Design, 10(6), 356–360. doi:10.1016/0010-4485(78)90111-2
- Dyn, N., & Levin, D. (2002). Subdivision schemes in geometric modelling. Acta Numerica, 11, 73–144. doi:10.1017/S0962492902000028
- Garg, A. (2020). Interactive, computation assisted design tools. Columbia University ProQuest Dissertations Publishing. 28026539.
- Hameed, R., & Mustafa, G. (2017). Family of a-point b-ary subdivision schemes with bell-shaped mask. Applied Mathematics and Computation, 309, 289–302. doi:10.1016/j.amc.2017.04.013
- Hassan, M. F., Ivrissimitzis, I. P., Dodgson, N. A., & Sabin, M. A. (2002). An interpolating 4-point C2 ternary stationary subdivision scheme. Computer Aided Geometric Design, 19(1), 1–18.
- Livesu, M. (2021). Scalable mesh refinement for canonical polygonal schemas of extremely high genus shapes. IEEE Transactions on Visualization and Computer Graphics, 27(1), 254–260. doi:10.1109/TVCG.2020.3010736
- Livesu, M., Pitzalis, L., & Cherchi, G. (2021). Optimal dual schemes for adaptive grid based hexmeshing. ACM Transactions on Graphics (TOG), 41(2), 1–14. doi:10.1145/3494456
- Mahalingam, D., & Koneru, S. S. (2020). Constructing an Interpolatory Subdivision Scheme from Doo-Sabin Subdivision, Project Report: MEC-572.
- Meng, Q. X., Xu, W. Y., Wang, H. L., Zhuang, X. Y., Xie, W. C., & Rabczuk, T. (2020). DigiSim—An open source software package for heterogeneous material modeling based on digital image processing. Advances in Engineering Software, 148, 102836. doi:10.1016/j.advengsoft.2020.102836
- Mustafa, G., & Rehman, N. A. (2010). The mask of (2b+4)-point n-ary subdivision scheme. Computing, 90, 1–14.
- Ning, X., Gong, K., Li, W., Zhang, L., Bai, X., & Tian, S. (2020). Feature refinement and filter network for person re-identification. IEEE Transactions on Circuits and Systems for Video Technology, 31(9), 3391–3402. doi:10.1109/TCSVT.2020.3043026
- Rehan, K., & Siddiqi, S. S. (2015). A combined binary 6-point subdivision scheme. Applied Mathematics and Computation, 270(C), 130–135. doi:10.1016/j.amc.2015.08.030
- Siddiqi, S. S., & Rehan, K. (2010). Improved binary four point subdivision scheme and new corner cutting scheme. Computers and Mathematics with Applications, 59, 2647–2657. doi:10.1016/j.camwa.2010.01.034
- Siddiqi, S. S., & Rehan, K. (2011). Curve subdivision schemes for geometric modelling. International Journal of Computer Mathematics, 88(4), 851–863. doi:10.1080/00207160.2010.482987
- Siddiqi, S. S., & Younis, M. (2013). Construction of m-point binary approximating subdivision schemes. Applied Mathematics Letters, 26(3), 337–343. doi:10.1016/j.aml.2012.09.016
- Thorne, T. (2021). Computer Graphics 17- Curves and Surfaces 2. cit, 04–11.
- Xumin, L., Xiaojun, W., Xianpeng, Y., & Cailing, L. (2013). Adaptive Doo-Sabin subdivision algorithm. Journal of Theoretical & Applied Information Technology, 49(1), 311–315.
- Zhang, L., Wu, J., Wang, T., Borji, A., Wei, G., & Lu, H. (2020). A multistage refinement network for salient object detection. IEEE Transactions on Image Processing, 29, 3534–3545. doi:10.1109/TIP.2019.2962688