UPEM Master 2 Informatique SIS. Digital Geometry. Topic 3: Discrete straight lines: definition and recognition. Yukiko Kenmochi.

Transcription

1 UPEM Master 2 Informatique SIS Digital Geometry Topic 3: Discrete straight lines: definition and recognition Yukiko Kenmochi October 12, 2016 Digital Geometry : Topic 3 1/30

2 Straight lines and their discretizations Straight line Definition (Straight line) A line in the Euclidean space R 2 is defined by where α, β, γ R. L = {(x, y) R 2 : αx + βy + γ = 0} We use a normalization, for example, α 2 + β 2 = 1, α + β = 1, or max( α, β ) = 1. Digital Geometry : Topic 3 2/30

3 Straight lines and their discretizations Discretization of straight line Definition (Discrete line) The discrete line of L in Z 2 is defined by D(L) = {(p, q) Z 2 : 0 αp + βq + γ ω} where ω is called the thickness. The values of γ and ω depend on the model of discretization. Digital Geometry : Topic 3 3/30

4 Straight lines and their discretizations Discretization of straight line Definition (Discrete line) The discrete line of L in Z 2 is defined by D(L) = {(p, q) Z 2 : 0 αp + βq + γ ω} where ω is called the thickness. The values of γ and ω depend on the model of discretization. Grid-intersection: γ = γ + ω 2, ω = max( α, β ). Digital Geometry : Topic 3 3/30

5 Straight lines and their discretizations Discretization of straight line Definition (Discrete line) The discrete line of L in Z 2 is defined by D(L) = {(p, q) Z 2 : 0 αp + βq + γ ω} where ω is called the thickness. The values of γ and ω depend on the model of discretization. Grid-intersection: γ = γ + ω 2, ω = max( α, β ). Super-cover (outer Jordan): γ = γ + ω 2, ω = α + β, Digital Geometry : Topic 3 3/30

6 Straight lines and their discretizations Discretization of straight line Definition (Discrete line) The discrete line of L in Z 2 is defined by D(L) = {(p, q) Z 2 : 0 αp + βq + γ ω} where ω is called the thickness. The values of γ and ω depend on the model of discretization. Grid-intersection: γ = γ + ω 2, ω = max( α, β ). Super-cover (outer Jordan): γ = γ + ω 2, ω = α + β, Gauss (half-plane): Digital Geometry : Topic 3 3/30

7 Straight lines and their discretizations Discretization of straight line Definition (Discrete line) The discrete line of L in Z 2 is defined by D(L) = {(p, q) Z 2 : 0 αp + βq + γ ω} where ω is called the thickness. The values of γ and ω depend on the model of discretization. Grid-intersection: γ = γ + ω 2, ω = max( α, β ). Super-cover (outer Jordan): γ = γ + ω 2, ω = α + β, Gauss (half-plane): γ = γ, ω detemines the m-connectedness of the half-plane border: ω = max( α, β ) 1 for m = 8, ω = α + β 1 for m = 4. Digital Geometry : Topic 3 3/30

8 Properties of discrete lines Freeman code Freeman code Chain code: Digital Geometry : Topic 3 4/30

9 Properties of discrete lines Properties of discrete lines Criteria of Freeman (1974) For discrete lines (by grid-intersection discretization), the Freeman code verifies the following three properties: Freeman code Discrete line : Digital Geometry : Topic 3 5/30

10 Properties of discrete lines Properties of discrete lines Criteria of Freeman (1974) For discrete lines (by grid-intersection discretization), the Freeman code verifies the following three properties: Freeman code Discrete line : Digital Geometry : Topic 3 5/30

11 Properties of discrete lines Properties of discrete lines Criteria of Freeman (1974) For discrete lines (by grid-intersection discretization), the Freeman code verifies the following three properties: 1 the code contains at most two different values; Freeman code Discrete line : Digital Geometry : Topic 3 5/30

12 Properties of discrete lines Properties of discrete lines Criteria of Freeman (1974) For discrete lines (by grid-intersection discretization), the Freeman code verifies the following three properties: 1 the code contains at most two different values; 2 those two values differ at most by one unit (modulo 8) ; Freeman code Discrete line : Digital Geometry : Topic 3 5/30

13 Properties of discrete lines Properties of discrete lines Criteria of Freeman (1974) For discrete lines (by grid-intersection discretization), the Freeman code verifies the following three properties: 1 the code contains at most two different values; 2 those two values differ at most by one unit (modulo 8) ; 3 one of the two values appears isolatedly and its appearances are uniformly spaced in the code. Freeman code Discrete line : Digital Geometry : Topic 3 5/30

14 Properties of discrete lines Properties of discrete lines (cont.) Definition (Chord property (Rosenfeld, 1974)) A set of discrete points X satisfies the chord property if for every pair of points p and q of X and for every point r = (r x, r y ) of the real segment between p and q, there exists a point s = (s x, s y ) of X such that max( s x r x, s y r y ) < 1. Digital Geometry : Topic 3 6/30

15 Properties of discrete lines Properties of discrete lines (cont.) Definition (Chord property (Rosenfeld, 1974)) A set of discrete points X satisfies the chord property if for every pair of points p and q of X and for every point r = (r x, r y ) of the real segment between p and q, there exists a point s = (s x, s y ) of X such that max( s x r x, s y r y ) < 1. It proves that a discrete curve is a discrete line segment if and only if it owns the chord property. Digital Geometry : Topic 3 6/30

16 Properties of discrete lines Properties of discrete lines (cont.) Definition (Chord property (Rosenfeld, 1974)) A set of discrete points X satisfies the chord property if for every pair of points p and q of X and for every point r = (r x, r y ) of the real segment between p and q, there exists a point s = (s x, s y ) of X such that max( s x r x, s y r y ) < 1. It proves that a discrete curve is a discrete line segment if and only if it owns the chord property. It allows to show the two first criteria of Freeman and to deduce a number of properties that specify the third criterion. Digital Geometry : Topic 3 6/30

17 Properties of discrete lines Properties of discrete lines (cont.) Definition (Chord property (Rosenfeld, 1974)) A set of discrete points X satisfies the chord property if for every pair of points p and q of X and for every point r = (r x, r y ) of the real segment between p and q, there exists a point s = (s x, s y ) of X such that max( s x r x, s y r y ) < 1. It proves that a discrete curve is a discrete line segment if and only if it owns the chord property. It allows to show the two first criteria of Freeman and to deduce a number of properties that specify the third criterion. There are a number of algorithms for recognizing a discrete straight line based on this property. Digital Geometry : Topic 3 6/30

18 Discrete line drawing Bresenham line-drawing algorithm Algorithm: drawing a discrete line (Bresenham, 1962) Input: Two discrete points (x 1, y 1 ), (x 2, y 2 ) (s.t. x 2 x 1 y 2 y 1 > 0) Output: Line segment between the two points d x = x 2 x 1, d y = y 2 y 1 ; y = y 1 ; e = 0; for x from x 1 to x 2 do put the pixel (x, y); e = e + dy d x ; if e 1 then y = y + 1; e = e 1; initialization value of initial error Can we avoid fractional numbers (error accumulation)? Digital Geometry : Topic 3 7/30

19 Discrete line drawing Bresenham line-drawing algorithm Algorithm: drawing a discrete line (Bresenham, 1962) Input: Two discrete points (x 1, y 1 ), (x 2, y 2 ) (s.t. x 2 x 1 y 2 y 1 > 0) Output: Line segment between the two points d x = x 2 x 1, d y = y 2 y 1 ; y = y 1 ; e = 0; for x from x 1 to x 2 do put the pixel (x, y); e = e + d y ; if e d x then y = y + 1; e = e d x; initialization value of initial error Can we consider rounding instead of truncation? Digital Geometry : Topic 3 7/30

20 Discrete line drawing Bresenham line-drawing algorithm Algorithm: drawing a discrete line (Bresenham, 1962) Input: Two discrete points (x 1, y 1 ), (x 2, y 2 ) (s.t. x 2 x 1 y 2 y 1 > 0) Output: Line segment between the two points d x = x 2 x 1, d y = y 2 y 1 ; y = y 1 ; e = d x ; for x from x 1 to x 2 do put the pixel (x, y); e = e + 2d y ; if e 2d x then y = y + 1; e = e 2d x; initialization value of initial error All the calculations are realised by using integers. Digital Geometry : Topic 3 7/30

21 Arithmetic lines Arithmetic definition of discrete lines Definition (Arithmetic line) An arithmetic line of parameters (a, b, c) and of arithmetic thickness w is defined as D(a, b, c, w) = {(p, q) Z 2 : 0 ap + bq + c < w} where a, b, c, w Z and gcd(a, b) = 1. The thickness w allows to control the connectedness of the line. Digital Geometry : Topic 3 8/30

22 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: Digital Geometry : Topic 3 9/30

23 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: 1 if w < max( a, b ), it is not connected; 1 Digital Geometry : Topic 3 9/30

24 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: 1 if w < max( a, b ), it is not connected; 2 if w = max( a, b ), it is a 8-curve ; naive line 1 2 Digital Geometry : Topic 3 9/30

25 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: 1 if w < max( a, b ), it is not connected; 2 if w = max( a, b ), it is a 8-curve ; naive line 3 if max( a, b ) < w < a + b, it is a -curve (its two successive points are 4-adjacent or strictly 8-adjacent); 1 2 Digital Geometry : Topic 3 9/30

26 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: 1 if w < max( a, b ), it is not connected; 2 if w = max( a, b ), it is a 8-curve ; naive line 3 if max( a, b ) < w < a + b, it is a -curve (its two successive points are 4-adjacent or strictly 8-adjacent); 4 if w = a + b, it is a 4-curve; standard line Digital Geometry : Topic 3 9/30

27 Arithmetic lines Thickness and connectedness of arithmetic lines Theorem Let D(a, b, c, w) be an arithmetic line, then: 1 if w < max( a, b ), it is not connected; 2 if w = max( a, b ), it is a 8-curve ; naive line 3 if max( a, b ) < w < a + b, it is a -curve (its two successive points are 4-adjacent or strictly 8-adjacent); 4 if w = a + b, it is a 4-curve; standard line 5 if w > a + b, it is said thick Digital Geometry : Topic 3 9/30

28 Arithmetic lines Remainder and leaning point of arithmetic lines Definition (Remainder) The remainder associated to a point p = (p x, p y ) of D(a, b, c, w) is an integer value defined by R(p) = ap x + bp y + c. Digital Geometry : Topic 3 10/30

29 Arithmetic lines Remainder and leaning point of arithmetic lines Definition (Remainder) The remainder associated to a point p = (p x, p y ) of D(a, b, c, w) is an integer value defined by R(p) = ap x + bp y + c. When the remainder is 0, p is called a lower leaning point. When the remainder is w 1, p is called a upper leaning point. Digital Geometry : Topic 3 10/30

30 Arithmetic lines Remainder and leaning point of arithmetic lines Definition (Remainder) The remainder associated to a point p = (p x, p y ) of D(a, b, c, w) is an integer value defined by R(p) = ap x + bp y + c. When the remainder is 0, p is called a lower leaning point. When the remainder is w 1, p is called a upper leaning point. We can generalize the Bresenham algorithm by using the remainder instead of the error e. Digital Geometry : Topic 3 10/30

31 Arithmetic lines Arithmetic line drawing algorithm Algorithm: drawing an arithmetic (naive) line Input: Two discrete points (x 1, y 1 ), (x 2, y 2 ) and c Output: Line segment between the two points b = x 2 x 1, a = y 2 y 1 ; y = y 1 ; r = ax 1 + by 1 + c; for x from x 1 to x 2 put the pixel (x, y); r = ; if then y = y + 1; r = ; We consider here that x 2 x 1 y 2 y 1 > 0. The value of c is initially chosen such that 0 ax 1 + by 1 + c < b. Digital Geometry : Topic 3 11/30

32 Arithmetic lines Arithmetic line drawing algorithm Algorithm: drawing an arithmetic (naive) line Input: Two discrete points (x 1, y 1 ), (x 2, y 2 ) and c Output: Line segment between the two points b = x 2 x 1, a = y 2 y 1 ; y = y 1 ; r = ax 1 + by 1 + c; for x from x 1 to x 2 put the pixel (x, y); r = r + a; if r b then y = y + 1; r = r b; We consider here that x 2 x 1 y 2 y 1 > 0. The value of c is initially chosen such that 0 ax 1 + by 1 + c < b. Digital Geometry : Topic 3 11/30

33 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? Digital Geometry : Topic 3 12/30

34 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? There are many recognition algorithms with linear complexity. Digital Geometry : Topic 3 12/30

35 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? There are many recognition algorithms with linear complexity. 1 approach of linear programming (Werman et al, 1987): verify the existence of feasible (real) solutions. Digital Geometry : Topic 3 12/30

36 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? There are many recognition algorithms with linear complexity. 1 approach of linear programming (Werman et al, 1987): verify the existence of feasible (real) solutions. 2 linguistic approach (Smeulders, Dorst, 1991): use the linguistic properties of discrete lines. Digital Geometry : Topic 3 12/30

37 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? There are many recognition algorithms with linear complexity. 1 approach of linear programming (Werman et al, 1987): verify the existence of feasible (real) solutions. 2 linguistic approach (Smeulders, Dorst, 1991): use the linguistic properties of discrete lines. 3 approach based on preimage (Lindenbaum, Bruckstein, 1993): use the properties of discrete lines in the dual space, called preimages. Digital Geometry : Topic 3 12/30

38 Discrete line recognition Discrete line recognition Problem (Discrete line recognition) Given a discrete curve X (more generally, a set of discrete points), is there a discrete line such that all the points of X belong to it? Yes or No If yes, what are the parameters of this discrete line? There are many recognition algorithms with linear complexity. 1 approach of linear programming (Werman et al, 1987): verify the existence of feasible (real) solutions. 2 linguistic approach (Smeulders, Dorst, 1991): use the linguistic properties of discrete lines. 3 approach based on preimage (Lindenbaum, Bruckstein, 1993): use the properties of discrete lines in the dual space, called preimages. 4 arithmetic approach (Debled-Rennesson, Reveillès, 1995): verify the existence of integer solutions by using arithmetic properties. Digital Geometry : Topic 3 12/30

39 Discrete line recognition Incremental arithmetic line recognition: initialization Let S be a segment of naive line D(a, b, c) with 0 a < b, q = (xq, yq) be the point of the greatest abscissa of S, l and l be the lower leaning points of minimum and maximum abscissas of S, u and u be the upper leaning points of minimum and maximum abscissas of S. By adding a point p = (xp, yp) connected to S such that xp = xq + 1, we verify if S = S {p} is still a naive line segment. Digital Geometry : Topic 3 13/30

40 Discrete line recognition Incremental arithmetic line recognition: algorithm Theorem (Debled-Rennesson and Reveillès, 1995) We have 1 if 0 r(p) < b, S is a naive line segment D(a, b, c); 1 Digital Geometry : Topic 3 14/30

41 Discrete line recognition Incremental arithmetic line recognition: algorithm Theorem (Debled-Rennesson and Reveillès, 1995) We have 1 if 0 r(p) < b, S is a naive line segment D(a, b, c); 2 if r(p) < 1 or r(p) > b, then S is not a naive line segment; 1 Digital Geometry : Topic 3 14/30

42 Discrete line recognition Incremental arithmetic line recognition: algorithm Theorem (Debled-Rennesson and Reveillès, 1995) We have 1 if 0 r(p) < b, S is a naive line segment D(a, b, c); 2 if r(p) < 1 or r(p) > b, then S is not a naive line segment; 3 if r(p) = 1, then S is a naive line segment D(yp yu, xp xu, axp + byp); 1 Digital Geometry : Topic 3 14/30

43 Discrete line recognition Incremental arithmetic line recognition: algorithm Theorem (Debled-Rennesson and Reveillès, 1995) We have 1 if 0 r(p) < b, S is a naive line segment D(a, b, c); 2 if r(p) < 1 or r(p) > b, then S is not a naive line segment; 3 if r(p) = 1, then S is a naive line segment D(yp yu, xp xu, axp + byp); 4 if r(p) = b, then S is a naive line segment D(yp y l, xp x l, axp + byp + b 1). 1 4 Digital Geometry : Topic 3 14/30

44 Discrete line recognition Farey sequence Definition (Farey sequence (Hardy and Write, 1979)) The Farey sequence F n of order n is the sequence of irreducible fractions between 0 and 1, whose denominators are less than or equal to n, in ascending order. Digital Geometry : Topic 3 15/30

45 Discrete line recognition Farey sequence Definition (Farey sequence (Hardy and Write, 1979)) The Farey sequence F n of order n is the sequence of irreducible fractions between 0 and 1, whose denominators are less than or equal to n, in ascending order. If 0 h k n and gcd(h, k) = 1, then h k is in F n. Digital Geometry : Topic 3 15/30

46 Discrete line recognition Farey sequence Definition (Farey sequence (Hardy and Write, 1979)) The Farey sequence F n of order n is the sequence of irreducible fractions between 0 and 1, whose denominators are less than or equal to n, in ascending order. If 0 h k n and gcd(h, k) = 1, then h k is in F n. Example (F 5 ) 0 1, 1 5, 1 4, 1 3, 2 5, 1 2, 3 5, 2 3, 3 4, 4 5, 1 1 Digital Geometry : Topic 3 15/30

47 Discrete line recognition Structure of the Farey sequence: Stern-Brocot tree Property (Neighborhood) If a b and c d are neighboring in a Farey sequence, with a b < c d, then their difference is equal to 1 bd. Digital Geometry : Topic 3 16/30

48 Discrete line recognition Structure of the Farey sequence: Stern-Brocot tree Property (Neighborhood) If a b and c d are neighboring in a Farey sequence, with a b < c d, then their difference is equal to 1 bd. Property (Median) If a b, p q and c d are neighboring in a Farey sequence such that a b < p q < c d, then p q is the median of a b and c d such as p q = a + c b + d. Digital Geometry : Topic 3 16/30

49 Discrete line recognition Structure of the Farey sequence: Stern-Brocot tree Property (Neighborhood) If a b and c d are neighboring in a Farey sequence, with a b < c d, then their difference is equal to 1 bd. Property (Median) If a b, p q and c d are neighboring in a Farey sequence such that a b < p q < c d, then p q is the median of a b and c d such as p q = a + c b + d. These properties allow to construct the Stern-Brocot tree. Digital Geometry : Topic 3 16/30

50 Discrete line recognition Stern-Brocot tree and discrete lines Each vertex h k of the tree corresponds to a pattern (motif) associated to the discrete line of slope h k. Digital Geometry : Topic 3 17/30

51 Discrete line recognition Stern-Brocot tree and discrete lines Each vertex h k of the tree corresponds to a pattern (motif) associated to the discrete line of slope h k. Updating parameters of the incremental discrete line recognition algorithm indicates moving from the tree root to a leaf. (Debled-Rennesson, 1995) Digital Geometry : Topic 3 17/30

52 Discrete line recognition Preimage of a discrete line segment Definition (Preimage of a discrete line segment (Anderson, Kim, 1985)) Given a discrete line segment S, the preimage P(S) is defined as the set of all lines y = sx + t having the same discretization S. Each point (p i, q i ) S, i = 1, 2,..., S, satisfies 0 sp i + q i + t < 1. Thus, P(S) in the parameter space (s, t) is the intersection of 2 S half-planes or S strips of height 1. y t s x Discrete line segment Preimage Digital Geometry : Topic 3 18/30

53 Discrete line recognition Properties of preimages and incremental discrete line recognition Property (Preimage (Dorst, Smeulders, 1984)) Any preimage P(S) forms a convex polygon with at most four vertices in the parameter space. If P(S) has four vertices, then two of them have a common s-coordinate, which is between the s-coordinates of the other two vertices. Property (Preimage (McIlroy, 1984)) The s-coordinates of the vertices of P(S) are successive terms of the Farey sequence of order max(p 0, S p 0 ) where p 0 is the minimal x-coordinate of S. Digital Geometry : Topic 3 19/30

54 Discrete line recognition Properties of preimages and incremental discrete line recognition Property (Preimage (Dorst, Smeulders, 1984)) Any preimage P(S) forms a convex polygon with at most four vertices in the parameter space. If P(S) has four vertices, then two of them have a common s-coordinate, which is between the s-coordinates of the other two vertices. Property (Preimage (McIlroy, 1984)) The s-coordinates of the vertices of P(S) are successive terms of the Farey sequence of order max(p 0, S p 0 ) where p 0 is the minimal x-coordinate of S. These properties allow to improve the incremental algorithm for discrete line recognition with optimal complexity in linear time and constant space. For checking if the set S = S {(p, q)} is a discrete line segment, we verify if P(S ) = P(S) {(s, t) 0 sp + q + t < 1} is not empty. (Lindenbaum, Bruckstein, 1993) Digital Geometry : Topic 3 19/30

55 Discrete line recognition Applications of discrete line recognition The discrete line recognition allow us to: Digital Geometry : Topic 3 20/30

56 Discrete line recognition Applications of discrete line recognition The discrete line recognition allow us to: study the parallelism, colinearity, orthogonality, convexity in the discrete space; Digital Geometry : Topic 3 20/30

57 Discrete line recognition Applications of discrete line recognition The discrete line recognition allow us to: study the parallelism, colinearity, orthogonality, convexity in the discrete space; estimate geometric properties of discrete object borders, such as the length of a curve, tangent and curvature at a point in a curve, etc.; Digital Geometry : Topic 3 20/30

58 Discrete line recognition Applications of discrete line recognition The discrete line recognition allow us to: study the parallelism, colinearity, orthogonality, convexity in the discrete space; estimate geometric properties of discrete object borders, such as the length of a curve, tangent and curvature at a point in a curve, etc.; make a segmentation of a discrete curve into line segments (polygonal approximation). Digital Geometry : Topic 3 20/30

59 Discrete line recognition Applications of discrete line recognition The discrete line recognition allow us to: study the parallelism, colinearity, orthogonality, convexity in the discrete space; estimate geometric properties of discrete object borders, such as the length of a curve, tangent and curvature at a point in a curve, etc.; make a segmentation of a discrete curve into line segments (polygonal approximation). If there is noise in discrete object border, we need to modify the problem. Digital Geometry : Topic 3 20/30

60 Discrete line fitting Discrete line fitting Problem (Discrete line fitting) Given a finite set of discrete points such that S = {p i Z 2 : i = 1, 2,..., N}, we seek the maximum subset of S whose elements are contained by some discrete line D(L). Digital Geometry : Topic 3 21/30

61 Discrete line fitting Discrete line fitting Problem (Discrete line fitting) Given a finite set of discrete points such that S = {p i Z 2 : i = 1, 2,..., N}, we seek the maximum subset of S whose elements are contained by some discrete line D(L). Points p i S are called inliers if p i S D(L); otherwise, they are called outliers. Then, the problem is also described as finding the maximum inlier set in S, called the optimal consensus of discrete line fitting. Digital Geometry : Topic 3 21/30

62 Discrete line fitting Deterministic method for discrete line fitting In the line parameter space, each point p i S corresponds to the strip of height 1. Thus the set of all the strips of S divides the parameter space into regions (as preimages). The problem is to find the region where the maximum number of strips pass. Solution (Kenmochi et al. 2010) By using the topological sweep algorithm (for arrangement of lines), we can visit all the regions with the time complexity O(N 2 ) and the space complexity O(N) where N is the number of points. Digital Geometry : Topic 3 22/30

63 Discrete line fitting Probabilistic method for discrete line fitting (RANSAC) If you know the probability information of data (you know how many outliers are contained approximately), then you can try RANSAC (RANdom Sample Consensus). A solution is obtained with a certain probability; if we increase this probability, we need more iterations, whose number can be fixed by the above probability information of data. RANSAC (Fischler, Bolles, 1981) Iteratively, 1 choose two points, and make a discrete line model from them; 2 count the number of inliers, and compare it to the current best one; 3 if it is better, then replace it as the current best. Alternative robust line fitting: Hough transform Digital Geometry : Topic 3 23/30

64 3D discrete lines 3D straight lines and their discretization Definition (3D straight line) A straight line in the Euclidean space R 3 is defined by L = {(α 1 t + β 1, α 2 t + β 2, α 3 t + β 3 ) R 3 : t R} where α i, β i R for i = 1, 2, 3. Digital Geometry : Topic 3 24/30

65 3D discrete lines 3D straight lines and their discretization Definition (3D straight line) A straight line in the Euclidean space R 3 is defined by L = {(α 1 t + β 1, α 2 t + β 2, α 3 t + β 3 ) R 3 : t R} where α i, β i R for i = 1, 2, 3. The discrete line D(L) defined in Z 3 by the grid intersection is the set of discrete points that are closest to the intersection in the plane of the grid. Digital Geometry : Topic 3 24/30

66 3D discrete lines Discretized line and discrete curve A discretized line is a 26-curve. Definition (m-curve) An m-path π is an m-curve if for every element p i of π, i = 1,..., n, p i has exactly two m-adjacent points in π, except for p 0 and p n that has only one. Digital Geometry : Topic 3 25/30

67 3D discrete lines Discretized line and discrete curve A discretized line is a 26-curve. Theorem (Kim, 1983) A 26-curve is a discrete line if and only if two of its projections on the xy-, yz- and zx-planes are 8-connected 2D discrete lines. Digital Geometry : Topic 3 25/30

68 3D discrete lines Discretized line and discrete curve A discretized line is a 26-curve. Theorem (Kim, 1983) A 26-curve is a discrete line if and only if two of its projections on the xy-, yz- and zx-planes are 8-connected 2D discrete lines. The 3D discrete line recognition is realized by the 2D discrete line recognition (three times). Digital Geometry : Topic 3 25/30

69 3D discrete lines Arithmetic definition of 3D discrete lines Definition (3D arithmetic line) A set G Z 3 is an arithmetic line defined by seven integer parameters a, b, c, d 1, d 2, w 1, and w 2 if and only if G = {(x, y, z) Z 3 : d 1 cx az < d 1 +w 1 d 2 bx ay < d 2 +w 2 }. For simplification, we consider 0 c b a and gcd(a, b, c) = 1. Digital Geometry : Topic 3 26/30

70 3D discrete lines Arithmetic definition of 3D discrete lines Definition (3D arithmetic line) A set G Z 3 is an arithmetic line defined by seven integer parameters a, b, c, d 1, d 2, w 1, and w 2 if and only if G = {(x, y, z) Z 3 : d 1 cx az < d 1 +w 1 d 2 bx ay < d 2 +w 2 }. For simplification, we consider 0 c b a and gcd(a, b, c) = 1. The parameters d 1 and d 2 are called the lower bounds and the parameters w 1 and w 2 define the arithmetic thickness Digital Geometry : Topic 3 26/30

71 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: Digital Geometry : Topic 3 27/30

72 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: 1 if a + c w 1 and a + b w 2, G is 6-connected; Digital Geometry : Topic 3 27/30

73 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: 1 if a + c w 1 and a + b w 2, G is 6-connected; 2 if a + c w 1 and a w 2 < a + b, or if a + b w 2 and a w 1 < a + c, G is 18-connected; Digital Geometry : Topic 3 27/30

74 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: 1 if a + c w 1 and a + b w 2, G is 6-connected; 2 if a + c w 1 and a w 2 < a + b, or if a + b w 2 and a w 1 < a + c, G is 18-connected; 3 if a w 1 < a + c and a w 2 < a + b, G is 26-connected; Digital Geometry : Topic 3 27/30

75 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: 1 if a + c w 1 and a + b w 2, G is 6-connected; 2 if a + c w 1 and a w 2 < a + b, or if a + b w 2 and a w 1 < a + c, G is 18-connected; 3 if a w 1 < a + c and a w 2 < a + b, G is 26-connected; 4 if w 1 < a or w 2 < a, G is not connected. Digital Geometry : Topic 3 27/30

76 3D discrete lines Thickness and connectedness of 3D discrete line The thickness w 1 and w 2 allow to control the connectedness of the line. Theorem (Coeurjolly et al., 2001) Let G be a discrete line defined by a, b, c, d 1, d 2, w 1, w 2 Z where 0 c < b < a, then: 1 if a + c w 1 and a + b w 2, G is 6-connected; 2 if a + c w 1 and a w 2 < a + b, or if a + b w 2 and a w 1 < a + c, G is 18-connected; 3 if a w 1 < a + c and a w 2 < a + b, G is 26-connected; 4 if w 1 < a or w 2 < a, G is not connected. G is called a 3D naive line if and only if w 1 = w 2 = max( a, b, c ). Digital Geometry : Topic 3 27/30

77 3D discrete lines 3D naive line Theorem (Coeurjolly et al., 2001) A rational line discretized by the grid intersection is a 3D naive line and vice-versa. Digital Geometry : Topic 3 28/30

78 3D discrete lines 3D naive line Theorem (Coeurjolly et al., 2001) A rational line discretized by the grid intersection is a 3D naive line and vice-versa. According to Theorem (Kim, 1983), we obtain the following corollary: Corollary (Coeurjolly et al., 2001) A 26-curve is a 3D naive line if and only if two of its projections on the xy-, yz- and zx-planes are 2D naives lines. Digital Geometry : Topic 3 28/30

79 3D discrete lines 3D discrete line recognition Problem (3D discrete line recognition) Given a set of 3D discrete points X, do the points of X belong to a 3D discrete line? Yes or No If yes, what are the parameters of this discrete line? Digital Geometry : Topic 3 29/30

80 3D discrete lines 3D discrete line recognition Problem (3D discrete line recognition) Given a set of 3D discrete points X, do the points of X belong to a 3D discrete line? Yes or No If yes, what are the parameters of this discrete line? We apply the incremental algorithm for arithmetic line recognition (for a 2D naive line) (Debled-Rennesson and Reveillès, 1995) to each projection of X on the xy-, yz- and zx-planes. Digital Geometry : Topic 3 29/30

81 3D discrete lines 3D discrete line recognition Problem (3D discrete line recognition) Given a set of 3D discrete points X, do the points of X belong to a 3D discrete line? Yes or No If yes, what are the parameters of this discrete line? We apply the incremental algorithm for arithmetic line recognition (for a 2D naive line) (Debled-Rennesson and Reveillès, 1995) to each projection of X on the xy-, yz- and zx-planes. If Yes for two of its projections, then Yes. Digital Geometry : Topic 3 29/30

82 References References I. Sivignon et I. Debled-Rennesson. Droites et plans discrets, Chapitre 6 dans Géométrie discrète et images numériques, Hermès, R. Klette and A. Rosenfeld. 2D Straightness, Chapter 9 in Digital geometry: geometric methods for digital picture analysis, Morgan Kaufmann, Digital Geometry : Topic 3 30/30