bboxerase

Remove bounding boxes

Since R2021a

Syntax

bboxB = bboxerase(bboxA,window)

[bboxB,indices] = bboxerase(bboxA,window)

___ = bboxerase(___,EraseThreshold=threshold)

Description

bboxB = bboxerase(bboxA,window) removes bounding boxes in the input bboxA that lie within a region of interest (ROI) specified by window. The output is the set of bounding boxes retained from the input bboxA. This function supports 2-D and 3-D bounding boxes.

Note

To perform random erase or cutout data augmentation, use bboxerase along with the imerase function.

[bboxB,indices] = bboxerase(bboxA,window) also returns the indices of the bounding boxes retained from the input set of bounding boxes bboxIn.

___ = bboxerase(___,EraseThreshold=threshold) additionally specifies the threshold for the amount of overlap between a bounding box region and the specified ROI. A bounding box is removed if the overlap between the bounding box region and the ROI is equal to or greater than the specified threshold.

Examples

Perform Random Erase Augmentation

Read an image.

I = imread('peppers.png');

Define bounding boxes and labels.

bboxA = [410 230 100 90;
        186 78 80 60];
labelsA = ["garlic";"onion"];

Find the size of the input image

inputSize = size(I);

To randomly select a region of interest (ROI), specify a range for the scale and the aspect ratio of the ROI. The scale value for the ROI is set to lie in the range 0.2 and 0.3. Similarly, the minimum value for the aspect ratio is set to lie in the range 1:10 and the maximum value for the aspect ratio is set to lie in the range 30:100.

scale = [0.2,0.3];
dimensionRatio = [1,10;30,100];

Select the ROI by using the randomWindow2d function.

window = randomWindow2d(inputSize,'Scale',scale,'DimensionRatio',dimensionRatio);

Remove the pixels and the bounding boxes that lie within the randomly selected ROI.

J = imerase(I,window);
[bboxB,indices] = bboxerase(bboxA,window);
labelsB = labelsA(indices);

Display the original and augmented image.

annotatedI = insertObjectAnnotation(I,'Rectangle',bboxA,labelsA);
annotatedJ = insertObjectAnnotation(J,'Rectangle',bboxB,labelsB);
figure
montage({annotatedI,annotatedJ})
title(['Input | Random Erase Output'])

Perform Cutout Data Augmentation

Read an image.

I = imread('visionteam1.jpg');

Define bounding boxes and labels.

bboxA = [64 101 117 440;
193 101 67 309;
282 86 114 375;
618 118 79 345;
486 55 131 528;
475 109 66 361];

labelsA = ["Person 1";"Person 2";"Person 3";"Person 4";"Person 5";"Person 6"];

Specify a rectangular region of interest.

window = [470 100 90 360];

Erase the pixels that lie within the rectangular region by using imerase function.

J = imerase(I,window);

Erase the bounding boxes that lie within the rectangular region. Set the erase threshold value to 0.7. The bboxerase function returns the retained bounding boxes and the corresponding indices.

[bboxB,indices] = bboxerase(bboxA,window,'EraseThreshold',0.7);

Read the class labels corresponding to the retained bounding boxes.

labelsB = labelsA(indices);

Display the results.

figure
I = insertObjectAnnotation(I,'Rectangle',bboxA,labelsA);
J = insertObjectAnnotation(J,'Rectangle',bboxB,labelsB);
imshowpair(I,J,'montage')
title(['Number of input bounding boxes: ', num2str(length(labelsA)), ' | ', 'Number of output bounding boxes: ', num2str(length(labelsB))])

Input Arguments

bboxA — Input set of bounding boxes

M-by-4 matrix

Input set of bounding boxes, specified as M-by-4 matrix. M is the number of bounding boxes. Each row in the matrix of form [x_min y_min width height]. [x_min y_min] are the top left coordinates of the bounding box. width and height are the width and the height of the bounding box respectively.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

window — Region of interest

4-element vector | Rectangle object

Region of interest, specified as one of these values:

• 4 -element vector of form [x y width height]. [x y] are the top left coordinates of the ROI. width and height are the width and the height of the ROI respectively.

• Rectangle object.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

threshold — Erase threshold

0.8 (default) | positive scalar

Erase threshold, specified as a positive scalar less than or equal to 1. The erase threshold sets the criteria for removing a bounding box from the input set bboxA. The value is the ratio of number of pixels in a bounding box that lie inside the ROI (overlapping pixels) to the total number of pixels in the bounding box.

$$threshold=\frac{Numberofoverlappingpixels}{Totalnumberofpixelsinboundingbox}$$

The default value is 0.8. This implies that, a bounding box is removed if 80 % of the pixels in the bounding box lie within the specified ROI.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Output Arguments

bboxB — Output bounding boxes retained from the input

N-by-4 matrix

Output bounding boxes retained from the input, returned as a N-by-4 matrix. N is the number of bounding boxes retained from the input. Each row of the matrix defines one bounding box of the same type as the input bboxA. The value of N is always less than or equal to M, the number of input bounding boxes.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

indices — Indices of retained bounding boxes

vector

Indices of retained bounding boxes, returned as a vector of integers. The indices indicate which bounding boxes in the input, bboxA, are retained and returned at the output.

Version History

Introduced in R2021a