bboxwarp

Apply geometric transformation to bounding boxes

Since R2019b

Syntax

bboxB = bboxwarp(bboxA,tform,ref)

bboxB = bboxwarp(bboxA,tform,ref,"OverlapThreshold",overlapThreshold)

[bboxB,indices] = bboxwarp(___)

Description

bboxB = bboxwarp(bboxA,tform,ref) transforms bounding boxes in bboxA according to the geometric transformation defined by tform. Bounding boxes can be axis-aligned rectangles, rotated rectangles, or cuboids. The spatial reference object, ref, defines the output view into which the boxes are transformed. This function supports 2-D and 3-D bounding boxes.

bboxB = bboxwarp(bboxA,tform,ref,"OverlapThreshold",overlapThreshold) also specifies the positive overlap threshold.

[bboxB,indices] = bboxwarp(___) additionally returns a vector of indices that indicate which bounding boxes in bboxA correspond to the warped versions in the output, bboxB. You can specify either of the preceding sets of input arguments.

Examples

Transform Images and Corresponding Bounding Boxes

Read an image.

I = imread("peppers.png");

Define bounding boxes and labels.

bboxA = [
    410 230 100 90
    186 78  80  60
    ]

bboxA = 2×4

   410   230   100    90
   186    78    80    60

labelsA = [
    "garlic"
    "onion"
    ];

Define an affine transform to horizontally flip and translate the image.

tform = affinetform2d([-1 0 50; 0 1 50; 0 0 1]);

Create an output view for imwarp.

rout = affineOutputView(size(I),tform);

Warp the image.

J = imwarp(I,tform,"OutputView",rout);

Warp the boxes.

[bboxB,indices] = bboxwarp(bboxA,tform,rout);
labelsB = labelsA(indices);

Display the results.

annotatedI = insertObjectAnnotation(I,"Rectangle",bboxA,labelsA);
annotatedJ = insertObjectAnnotation(J,"Rectangle",bboxB,labelsB);
figure
montage({annotatedI,annotatedJ})

Input Arguments

bboxA — Bounding boxes

M-by-4 matrix | M-by-5 matrix | M-by-9 matrix

Bounding boxes, specified as an M-by-4, M-by-5, or M-by-9 nonsparse numeric matrix of M bounding boxes. Each row, M, of the matrix defines a bounding box as either an axis-aligned rectangle, a rotate rectangle, or a cuboid. The table below describes the format of the bounding boxes.

Bounding Box	Description
Axis-aligned rectangle	Defined in spatial coordinates as an M-by-4 numeric matrix with rows of the form [x y w h], where: M is the number of axis-aligned rectangles. x and y specify the upper-left corner of the rectangle. w specifies the width of the rectangle, which is its length along the x-axis. h specifies the height of the rectangle, which is its length along the y-axis.
Rotated rectangle	Defined in spatial coordinates as an M-by-5 numeric matrix with rows of the form [xctr yctr xlen ylen yaw], where: M is the number of rotated rectangles. xctr and yctr specify the center of the rectangle. xlen specifies the width of the rectangle, which is its length along the x-axis before rotation. ylen specifies the height of the rectangle, which is its length along the y-axis before rotation. yaw specifies the rotation angle in degrees. The rotation is clockwise-positive around the center of the bounding box.
Cuboid	Defined in spatial coordinates as an M-by-9 numeric matrix with rows of the form [xctr yctr zctr xlen ylen zlen xrot yrot zrot], where: M is the number of cuboids. xctr, yctr, and zctr specify the center of the cuboid. xlen, ylen, and zlen specify the length of the cuboid along the x-axis, y-axis, and z-axis, respectively, before rotation. xrot, yrot, and zrot specify the rotation angles of the cuboid around the x-axis, y-axis, and z-axis, respectively. The xrot, yrot, and zrot rotation angles are in degrees about the cuboid center. Each rotation is clockwise-positive with respect to the positive direction of the associated spatial axis. The function computes rotation matrices assuming ZYX order Euler angles [xrot yrot zrot]. The figure shows how these values determine the position of a cuboid.

tform — Geometric transformation

affinetform2d object | affinetform3d object

Geometric transformation, specified as an affinetform2d object for rectangular inputs or an affinetform3d object for cuboid inputs. The bboxwarp function supports affine transformations consisting only of scale, rotation, and translation.

ref — Spatial reference

imref2d object | imref3d object

Spatial reference, specified as an imref2d object for rectangular inputs or imref3d object for cuboid inputs. To obtain one of these objects, you can use the imwarp or the affineOutputView function. The object defines the output view to transform boxes. Boxes that are transformed completely outside of the output view defined by ref are discarded.

[J,rout] = imwarp(I,tform);
[bboxB,indices] = bboxwarp(bboxA,tform,rout);

rout = affineOutputView(size(I),tform)
J = imwarp(I,tform,'OutputView',rout);
[bboxB,indices] = bboxwarp(bboxA,tform,rout);

overlapThreshold — Overlap threshold

1 (default) | positive scalar less than or equal to 1

Overlap threshold, specified as a positive scalar less than or equal to 1. The amount of overlap between transformed boxes and the region W, defined by the output view, is defined as:

area(intersect(bboxA,W))/area(bboxB,W).

If the computed overlap value is greater than the value of the threshold property, then the transformed boxes are clipped to the bounding rectangle border. Otherwise, the boxes are discarded. Lowering the threshold can result in parts of the object getting discarded.

Output Arguments

bboxB — Warped bounding boxes

M2-by-N matrix

Warped bounding boxes, returned as an M2-by-N matrix of M2 bounding boxes. The number of bounding boxes returned is less than the number of bounding boxes in the input. Each row, M2, of the matrix defines one bounding box of the same type as the input bboxA. When bboxB contains floating point data, the function returns it with the same type as bboxA. Otherwise, the function returns bboxB as type single.

indices — Indices

vector of integers

Indices, returned as a vector of integers. The indices indicate which bounding boxes in the input, bboxA, that correspond to the warped versions in the output, bboxB.

Version History

Introduced in R2019b

R2022b: Supports new geometric transformation objects

Starting in R2022b, most Computer Vision Toolbox™ functions create and perform geometric transformations using the premultiply convention. Accordingly, you can now specify tform as an affinetform2d or affinetform3d geometric transformation object, which use the premultiply convention.

Although you can still specify tform as an affine2d or affine3d geometric transformation object, these objects are not recommended because they use the postmultiply convention. You can streamline your geometric transformation workflows by switching to the new premultiply geometric transformation objects. For more information, see Migrate Geometric Transformations to Premultiply Convention.

R2022a: Bounding Box Coordinates: Data augmentation for object detection using spatial coordinates

The bboxresize, bboxcrop, bboxwarp, and showShape functions assume the input bounding box coordinates for axis-aligned rectangles are specified in spatial coordinates and return the transformed bounding boxes in spatial coordinates.