Is there any way to gain a smoother shape in binary image?

Question

I want to detect shapes like triangles, circles and rectangles in binary image after color segmentation but sometimes shapes are very ripped and jugged like in this picture. Is there any way to get a better shape?

Maybe that is something wrong with my color masks.

    def redColorDetection(self, img):

    self.low_red = np.array([160, 50, 5])
    self.high_red = np.array([180, 255, 255])

    red_mask = cv2.inRange(self.hsv, self.low_red, self.high_red)

    return  red_mask

those are manually cropped screenshots. you should supply the pictures themselves. for now this should suffice but keep it in mind for the future. — Christoph Rackwitz, Oct 14 '22 at 00:42
Is it right that your image is in HSV color space? Red color is around hue=0 and hue=180 (in opencv because 360 doesnt fit in 8 bit). One way to find why the wanted parts are not active in the mask is to READ the values from your hsv image to find out why they are not in the range of your inRange. — Micka, Oct 14 '22 at 11:47

score 5 · Answer 1 · answered Oct 14 '22 at 12:01

Red color is split in the cicular region of the HSV color space, so there is red part in 0+x and a red part in 360-x (180-x/2 in opencv because the 360° are divided by 2 there to fit into 8 bit).

For example 0+15 and 180-15 (+/-10 is also ok) in opencv give good results in your example:

int main()
{
    cv::Mat img = cv::imread("C:/data/StackOverflow/redSign.jpg");
    cv::Mat hsv;
    cv::cvtColor(img, hsv, cv::COLOR_BGR2HSV);

    cv::Scalar low_red1(165, 50, 5);
    cv::Scalar high_red1(180, 255, 255);

    cv::Scalar low_red2(0, 50, 5);
    cv::Scalar high_red2(15, 255, 255);

    cv::Mat mask1;
    cv::Mat mask2;
    cv::inRange(hsv, low_red1, high_red1, mask1);
    cv::inRange(hsv, low_red2, high_red2, mask2);

    cv::imshow("mask1", mask1);
    cv::imshow("mask2", mask2);

    cv::imshow("mask1+mask2", mask1 + mask2);

    cv::waitKey(0);

}

Giving this result:

Mask1:

Mask2:

Mask total:

If you want to segment red with a single inRange call you can shift the hue channel, but that's likely not more efficient, but might make it easier to write some kind of color segmentation general code: Have a look at the "shiftedH" part of: OpenCV Edge/Border detection based on color

score 5 · Answer 2 · answered Oct 14 '22 at 18:24

Here is another way to threshold on red by using the A channel from the LAB colorspace. The A channel is the red-green opponent colors, so red is bright and green is dark.

Input:

import cv2
import numpy as np

# load image
img = cv2.imread('sign.jpg')

# convert to LAB
lab = cv2.cvtColor(img,cv2.COLOR_BGR2LAB)

# extract A channel
# The A axis is relative to the green–red opponent colors
# See https://en.wikipedia.org/wiki/CIELAB_color_space
A = lab[:,:,1]

# threshold A channel (choose one or the other depending upon what result you want)
thresh1 = cv2.threshold(A, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)[1]
thresh2 = cv2.threshold(A, 150, 255, cv2.THRESH_BINARY)[1]

# save results
cv2.imwrite('sign_A_channel_lab.jpg', A)
cv2.imwrite('sign_threshold1.jpg', thresh1)
cv2.imwrite('sign_threshold2.jpg', thresh2)

# show results
cv2.imshow('A',A)
cv2.imshow('thresh1', thresh1)
cv2.imshow('thresh2', thresh2)
cv2.waitKey(0)
cv2.destroyAllWindows()

A channel from LAB:

Otsu Threshold:

Alternate Simple Threshold:

You could also require the B channel to be close to 0. That would remove the yellow and blue areas. — Cris Luengo, Oct 14 '22 at 18:56

Is there any way to gain a smoother shape in binary image?

2 Answers2