[image processing] pyefd.elliptic_ fourier_ How descriptors are used

Fourier descriptor , By means of Fourier series approximation , To approximate the representation of image contour .

The key input is contours, Image outline , Can pass opencv Function implementation of .

im = cv.cvtColor(input_imgs, cv.COLOR_BGR2GRAY)

_, thresh = cv.threshold(im, 254, 255, cv.THRESH_BINARY) #get binary

contours, hierarchy = cv.findContours(thresh,cv.RETR_TREE, cv.CHAIN_APPROX_SIMPLE )

The acquisition method of image contour is like the above code , Generally, it turns gray first , Get more binary, Send it to findContours Function .

Back to contours It's usually a four-dimensional vector , respectively 【 Number of contours , The number of contour points of the contour ,1, coordinate x and y】.

For each profile , Fourier descriptors can be used .

#!pip install pyefd
from pyefd import elliptic_fourier_descriptors, plot_efd, normalize_efd, reconstruct_contour
coeffs = []
for j in contours:
    
	coeffs.append(elliptic_fourier_descriptors(np.squeeze(j,  axis=1), order=10, normalize=True))

The dimension of each outline is 【 length ,1,2】, and elliptic_fourier_descriptors Can only handle 【 length ,2】 Data like this , So use np.squeeze().

Want to see the outline learned , It can be used plot_pyefd

plot_efd(coeffs[0])

order Is the number of images , It's also coeffs[0] The length of , And its corresponding width is always 4.

If you want to rebuild contour points ：

points = reconstruct_contour(coeffs)