Canny edge detector with threshold calculation using Otsu's method

detect_edges.py

import cv2
import numpy as np

def calculate_threshold(img):
	hist, bin_edges = np.histogram(img, bins=256)
	bin_mids = (bin_edges[:-1] + bin_edges[1:]) / 2.

	weight1 = np.cumsum(hist)
	weight2 = np.cumsum(hist[::-1])[::-1]
	mean1 = np.cumsum(hist * bin_mids) / weight1
	mean2 = (np.cumsum((hist * bin_mids)[::-1]) / weight2[::-1])[::-1]

	inter_class_variance = weight1[:-1] * weight2[1:] * (mean1[:-1] - mean2[1:]) ** 2
	index_of_max_val = np.argmax(inter_class_variance)

	return bin_mids[:-1][index_of_max_val]

def detect_edges(img):
	t = calculate_threshold(img)
	lower = int(max(0, t * 0.5))
	upper = int(min(255, t))

	edges = cv2.Canny(img, lower, upper)

	return edges

if __name__ == '__main__':
	img = cv2.imread("fish.jpg", cv2.IMREAD_GRAYSCALE)
	edges = 255 - detect_edges(img)

	cv2.imshow("Edges", edges)
	cv2.waitKey(0)