opencv for arbitrary shape target recognition, for your reference, is as follows
There was a time at work when I needed to perform target recognition on a simple graph, where the shape of the target was not fixed and there was some degree of noise influence, but the noise influence was indeterminate. It was a simple matter as the image was not complex and the code is now published below:
import cv2
def otsu_seg(img):
ret_th, bin_img = (img, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
return ret_th, bin_img
def find_pole(bin_img):
img, contours, hierarchy = (bin_img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
area = 0
for i in range(len(contours)):
area += (contours[i])
area_mean = area / len(contours)
mark = []
for i in range(len(contours)):
if (contours[i]) < area_mean:
(i)
return img, contours, hierarchy, mark
def draw_box(img,contours):
img = (img,
(contours[0][0], contours[0][1]),
(contours[1][0], contours[1][1]),
(255,255,255),
3)
return img
def main(img):
ret, th = otsu_seg(img)
img_new, contours, hierarchy, mark = find_pole(th)
for i in range(len(contours)):
if i not in mark:
left_point = contours[i].min(axis=1).min(axis=0)
right_point = contours[i].max(axis=1).max(axis=0)
img = draw_box(img, (left_point, right_point))
return img
if __name__ =="__main__":
img = ('G:/')
img = main(img)
('G:/test_d.png', img)
The resulting graph is shown below:
This is the whole content of this article.