Detect square in a List of Points

2.5k views Asked by At

I want to test if there is a square in a list of Point object or not.

This is my Point class :

class Point {
    private int x;
    private int y;

    public Point(int x, int y) {
        this.x = x;
        this.y = y;
    }

    public int getX() {
        return x;
    }

    public void setX(int x) {
        this.x = x;
    }

    public int getY() {
        return y;
    }

    public void setY(int y) {
        this.y = y;
    }

    public int distanceSquare(Point q) {
        return (x - q.getX()) * (x - q.getX()) +
                (y - q.getY()) * (y - q.getY());
    }
}

I can test if four Point is a Square or not :

static boolean isSquare(Point p1, Point p2, Point p3, Point p4) {
        int d2 = p1.distanceSquare(p2);  // from p1 to p2
        int d3 = p1.distanceSquare(p3);  // from p1 to p3
        int d4 = p1.distanceSquare(p4);  // from p1 to p4

        if (d2 == d3 && 2 * d2 == d4) {
            int d = p2.distanceSquare(p4);
            return (d == p3.distanceSquare(p4) && d == d2);
        }

        if (d3 == d4 && 2 * d3 == d2) {
            int d = p2.distanceSquare(p3);
            return (d == p2.distanceSquare(p4) && d == d3);
        }
        if (d2 == d4 && 2 * d2 == d3) {
            int d = p2.distanceSquare(p3);
            return (d == p3.distanceSquare(p4) && d == d2);
        }

        return false;
    }

But I didn't found the best way to search the square in a list of Point.

Can you help me !!!

3

There are 3 answers

10
leeyuiwah On BEST ANSWER

Updated to also detected rotated squares

I like Egor Skriptunoff's answer above, but let me try to give another answer. I think the complexity is only O(N^2).

Algorithm

For any pair of points (P0, P1) (there are N^2 of them), find out the vector V01 from P0 to P1, and then rotate this vector by 90 degrees (it becomes V12). Add it to P1, and see if you can find a point there? (this can be done by a hashmap lookup -- see more below) If so then you have P2, and continue the procedure.

Rotate the vector by another 90 degrees (it become V23), Add it to P2, and see if you can find a point there? (Again, by a hashmap lookup)
If so, then you have P3, and continue the procedure.

Rotate the vector by another 90 degrees (it becomes V34). Add it to P3, and see if can find a point there? (Again, by a hashmap lookup). If so, check also if this point P4 is the same as P0. If so, then you have just detected a square.

The following diagram illustrates the idea.

enter image description here

Data Structure

If the squares are rotatable, then the x and y coordinates of Point must be in floating point (double) and cannot be integer. Because a lot of computation will give you irrational numbers (e.g. sqrt(2))

However, a double representation cannot be precise as an integer, so we have to be careful to treat two double numbers that are close enough as the same double value. In my code, I use an epsilon as the tolerance that we allow for "approximate equivalence". I picked 1E-3 as the epsilon.

public class Point2 {
    private double x;
    private double y;
    private final double eps = 1E-3;    
    public double getEps() {
        return eps;
    }

And then in all computation regarding equals() and hashCode(), make sure you use a "snapped" value of x and y, not their original double representations. (Graphically you can imagine this like your graphical editor give you a "snap to grid" feature, with the grid size being epsilon). Also you need to be careful that in double reprsentations there are positive 0 and negative 0, and you should also treat them as the same value.

Something like this:

public long getXsnapped() {
    return Math.round(this.getX()/this.getEps());
}   
public long getYsnapped() {
    return Math.round(this.getY()/this.getEps());
}   
@Override
public int hashCode() {
    final int prime = 31;
    int result = 1;
    long temp;
    temp = Double.doubleToLongBits(eps);
    result = prime * result + (int) (temp ^ (temp >>> 32));
    if (Math.abs(this.getX())>this.getEps()) {
        // include X only if it is larger than eps
        temp = this.getXsnapped();
        result = prime * result + (int) (temp ^ (temp >>> 32));
    }
    if (Math.abs(this.getY())>this.getEps()) {
        temp = this.getYsnapped();
        result = prime * result + (int) (temp ^ (temp >>> 32));
    }
    return result;
}
@Override
public boolean equals(Object obj) {
    if (this == obj)
        return true;
    if (obj == null)
        return false;
    if (getClass() != obj.getClass())
        return false;
    Point2 other = (Point2) obj;
    if (Double.doubleToLongBits(eps) != Double.doubleToLongBits(other.eps))
        return false;
    boolean answer = true;
    if (Math.abs(this.getX())>this.getEps()
            || Math.abs(other.getX())>this.getEps()) {
        // compare value and sign only if X of both points are larger than eps
        if (this.getXsnapped()!= other.getXsnapped())
            answer = false;         
    }
    if (Math.abs(this.getY())>this.getEps()
            || Math.abs(other.getY())>this.getEps()) {
        // compare value and sign only if Y of both points are larger than eps
        if (this.getYsnapped()!= other.getYsnapped())
            answer &= false;
    }
    boolean isDebug = false;
    Util.debugPrint(isDebug, "p1 %s; p2 %s: %b (eps is %.9f)%n"
        , this, other, answer, this.getEps());
    return answer;
}

Furthermore, each square has four points. You can add a rule in your program saying that the four points must be in order as used in the algorithm (P0->P1->P2->P3) with the set angular relationship (see the algorithm above). However, you also need to be careful that the given the same four points there are four options to choose the start point. So your Square object code must treat the following squares specified by these four points as equivalent:

P0->P1->P2->P3
P1->P2->P3->P0 
P2->P3->P0->P1
P3->P0->P1->P2

This can be done in the constructor of the Square object and always "canonicalize" the input by picking a point with a certain sailent feature as the starting point (e.g. pick the point having the lowest x, and if its x value tie with another point, then pick the point having the lowest x and smaller y among the tied pair)

Test Input 1

-1.4142,    9.8995
-5.6569,   15.5563
1.4142,    9.8995
-1.4142,   14.1421
-2.1213,   14.8492
1.4142,   14.1421
0.0000,   15.5563
-2.1213,   17.6777
7.0711,   11.3137
5.6569,   12.7279
4.2426,   14.1421
6.3640,   10.6066
7.0711,   14.1421
5.6569,   15.5563
1.4142,   19.7990
7.7782,   14.8492

Test Result 1

===== Given a set of following points from file src\detectSquare\inputSet1_45_1_1_0_0.txt =====
1: Point2 [x=-1.4142, y=9.8995]
2: Point2 [x=-5.6569, y=15.5563]
3: Point2 [x=1.4142, y=9.8995]
4: Point2 [x=-1.4142, y=14.1421]
5: Point2 [x=-2.1213, y=14.8492]
6: Point2 [x=1.4142, y=14.1421]
7: Point2 [x=0.0000, y=15.5563]
8: Point2 [x=-2.1213, y=17.6777]
9: Point2 [x=7.0711, y=11.3137]
10: Point2 [x=5.6569, y=12.7279]
11: Point2 [x=4.2426, y=14.1421]
12: Point2 [x=6.3640, y=10.6066]
13: Point2 [x=7.0711, y=14.1421]
14: Point2 [x=5.6569, y=15.5563]
15: Point2 [x=1.4142, y=19.7990]
16: Point2 [x=7.7782, y=14.8492]
===== The following squares have been found =====
1: SquareRotatable [points=[Point2 [x=4.2427, y=14.1421], Point2 [x=5.6569, y=12.7279], Point2 [x=7.0711, y=14.1421], Point2 [x=5.6569, y=15.5563]]]

Test Input 2

0.0000,    0.0000
-0.7071,    0.7071
-1.4142,    1.4142
0.7071,    0.7071
0.0000,    1.4142
-0.7071,    2.1213
1.4142,    1.4142
0.7071,    2.1213
0.0000,    2.8284

Test Result 2

===== Given a set of following points from file src\detectSquare\inputSet2_45_0_0_0_0.txt =====
1: Point2 [x=0.0000, y=0.0000]
2: Point2 [x=-0.7071, y=0.7071]
3: Point2 [x=-1.4142, y=1.4142]
4: Point2 [x=0.7071, y=0.7071]
5: Point2 [x=0.0000, y=1.4142]
6: Point2 [x=-0.7071, y=2.1213]
7: Point2 [x=1.4142, y=1.4142]
8: Point2 [x=0.7071, y=2.1213]
9: Point2 [x=0.0000, y=2.8284]
===== The following squares have been found =====
1: SquareRotatable [points=[Point2 [x=-1.4142, y=1.4142], Point2 [x=0.0000, y=0.0000], Point2 [x=1.4142, y=1.4142], Point2 [x=0.0000, y=2.8284]]]
2: SquareRotatable [points=[Point2 [x=-0.7071, y=0.7071], Point2 [x=0.7071, y=0.7071], Point2 [x=0.7071, y=2.1213], Point2 [x=-0.7071, y=2.1213]]]
3: SquareRotatable [points=[Point2 [x=-1.4142, y=1.4142], Point2 [x=-0.7071, y=0.7071], Point2 [x=0.0000, y=1.4142], Point2 [x=-0.7071, y=2.1213]]]
4: SquareRotatable [points=[Point2 [x=-0.7071, y=2.1213], Point2 [x=0.0000, y=1.4142], Point2 [x=0.7071, y=2.1213], Point2 [x=-0.0000, y=2.8284]]]
5: SquareRotatable [points=[Point2 [x=-0.7071, y=0.7071], Point2 [x=0.0000, y=0.0000], Point2 [x=0.7071, y=0.7071], Point2 [x=0.0000, y=1.4142]]]
6: SquareRotatable [points=[Point2 [x=0.0000, y=1.4142], Point2 [x=0.7071, y=0.7071], Point2 [x=1.4142, y=1.4142], Point2 [x=0.7071, y=2.1213]]]

JUnit Test Program testing Point2#getXsnapped() (fragment only)

import org.junit.Before;
import org.junit.BeforeClass;
import org.junit.Ignore;
import org.junit.Test;

public class Point2Test {
    private List<Point2> points = new ArrayList<>();

    @Before
    public void setUp() throws Exception {
        points.add(new Point2(0.49999999f, 0));
        points.add(new Point2(0.50000001f, 0));
    }
    ...

    @Test
    public void testGetXsnapped() {
        System.out.format("testing xSnapped of two points: %s and %s%n"
                , points.get(0), points.get(1));
        System.out.format("and they are %d and %d respectively%n"
                , points.get(0).getXsnapped()
                , points.get(1).getXsnapped());
        System.out.format("(Note: epsilon is %f)%n"
                , points.get(0).getEps());

        assertEquals(points.get(0).getXsnapped()
                    , points.get(1).getXsnapped());
    }
}

Output of JUnit Test

testing xSnapped of two points: Point2 [x=0.5000, y=0.0000] and Point2 [x=0.5000, y=0.0000]
and they are 500 and 500 respectively
(Note: epsilon is 0.001000)

Caveat

Eric Duminil is correct in pointing out that there can be 2 points arbitrarily close to one another, and still be snapped to different points on the grid.

Off the top of my head, I don't know how to solve this problem. Suggestions are welcome!

E.g.

@Before
public void setUp() throws Exception {
    Point2 dummy = new Point2(0, 0);    // just to get epsilon
    points.add(new Point2(dummy.getEps()*0.5, 0));
    points.add(new Point2(dummy.getEps()*0.49999999999999, 0));
}

With these added debugging code:

public long getXsnapped() {
    boolean isDebug = true;
    String _ = "    "; // indent
    double X = this.getX();
    Util.debugPrint(isDebug, _ + "X is %E (long bits is 0x%x)%n"
                                , X, Double.doubleToLongBits(X));
    double eps = this.getEps();
    Util.debugPrint(isDebug, _ + "eps is %E%n", eps);
    double fraction = X/eps;
    Util.debugPrint(isDebug, _ + "fraction is %E (long bits is 0x%x)%n"
                                , fraction, Double.doubleToLongBits(fraction));
    long answer = Math.round(fraction); 
    Util.debugPrint(isDebug, _ + "xSnapped is %d%n", answer);
    return answer;
}   

Util.debugPrint():

public static void debugPrint(boolean isDebug, String format, Object... args) {
    if (!isDebug) {
        return; // don't print
    }
    System.out.format(format, args);
}

I would get following output -- the two points are considered different!

testing xSnapped of two points: Point2 [x=0.0005, y=0.0000] and Point2 [x=0.0005, y=0.0000]
    X is 5.000000E-04 (long bits is 0x3f40624dd2f1a9fc)
    eps is 1.000000E-03
    fraction is 5.000000E-01 (long bits is 0x3fe0000000000000)
    xSnapped is 1
    X is 5.000000E-04 (long bits is 0x3f40624dd2f1a9a0)
    eps is 1.000000E-03
    fraction is 5.000000E-01 (long bits is 0x3fdfffffffffff4c)
    xSnapped is 0
and they are 1 and 0 respectively
(Note: epsilon is 0.001000)
delta between the two x (as double) is 9.974660E-18
    X is 5.000000E-04 (long bits is 0x3f40624dd2f1a9fc)
    eps is 1.000000E-03
    fraction is 5.000000E-01 (long bits is 0x3fe0000000000000)
    xSnapped is 1
    X is 5.000000E-04 (long bits is 0x3f40624dd2f1a9a0)
    eps is 1.000000E-03
    fraction is 5.000000E-01 (long bits is 0x3fdfffffffffff4c)
    xSnapped is 0
0
Yola On

Here is O(n^2 lg(n))-time and O(n)-space algorithm. If you will presort vertices by angle, then for every pair {A,B} you can find both angles OC and OD.

compute angle for every point -- O(n)
sort point by angles -- O(n*lg(n))
for each pair -- O(n^2*lg(n))
    if (B.x > A.x) && (B.y > A.y)
        compute positions of C and D
        compute angles of C and D
        for C and D
            if there is a pair of points with same angles and same positions
                return true
return false

enter image description here

How to find coordinates of C:

Xc = Xb - (Yb - Ya)
Yc = Yb + (Xb - Xa)
8
Egor Skriptunoff On

Take all pairs of points; for each pair calculate its angle, its length and the middle point. This will take O(n^2) time.
For each set of pairs having the same middle point and the same length sort these pairs by angle, this will take in total O(n^2*log(n)) time. This will help you to find two orthogonal diagonals of the same length having the same middle point (that is the square!).
Total algo time complexity is O(n^2*log(n)).