Avoid bring images and over-exposed images. Corrects them and finds proper shapes.

bright_correction.py

import cv2
import numpy as np
from matplotlib import pyplot as plt


def morph_close(img):
	kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5,5))
	o = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)
	c = cv2.Canny(o,100,150)
	#div = np.float32(img)/o
	res = np.uint8(cv2.normalize(c, c, 0, 255,cv2.NORM_MINMAX))
	return c

img_col = cv2.imread('bright_numb.jpg')
img = cv2.cvtColor(img_col, cv2.COLOR_RGB2GRAY)

flag = True

if flag:

	equ = cv2.equalizeHist(img)
	clahe = cv2.createCLAHE(clipLimit=3.0, tileGridSize=(8,8))
	cl1 = clahe.apply(img)

	kernel = np.ones((5,5),np.uint8)
	gradient = cv2.morphologyEx(img, cv2.MORPH_GRADIENT, kernel)
	#erosion = cv2.erode(img,kernel,iterations = 1)
	c = cv2.Canny(cl1,100,200)
	t = cv2.adaptiveThreshold(c,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,11,2)

else:
	t = morph_close(img)
	#t = cv2.adaptiveThreshold(img,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY,11,2)

image, contours, hierarchy = cv2.findContours(t, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)


for cnt in contours:
	epsilon = 0.01*cv2.arcLength(cnt, False)
	approx = cv2.approxPolyDP(cnt, epsilon, False)
	area = cv2.contourArea(approx)
	rect = cv2.minAreaRect(approx)
	box = cv2.boxPoints(rect)
	box = np.int0(box)
	percentage = (area * 100) / (img.shape[0] * img.shape[1])
	if percentage > 0.2 and percentage < 20:
		cv2.drawContours(img_col,[box],0,(0,255,0),20)

cv2.imwrite('results/morph.jpg',t)
cv2.imwrite('results/shaped.jpg',img_col)