Skip to content

Instantly share code, notes, and snippets.

What would you like to do?
Code for Bangla Character Segmentation (Modified from avidLearnerInProgress/sudoku-solver-openCV-python)
#Author: Md Mahedi Hasan
#Date : 10/02/2019
#Description: Code for character segmentation in image
import cv2
import operator
import numpy as np
from matplotlib import pyplot as plt
def show_image(img):
cv2.imshow('image', img) #Display the image
cv2.waitKey(0) #Wait for any key to be pressed (with the image window active)
cv2.destroyAllWindows() #Close all windows
def plot_many_images(images, titles, rows=1, columns=2):
for i, image in enumerate(images):
plt.subplot(rows, columns, i+1)
plt.imshow(image, 'gray')
plt.xticks([]), plt.yticks([]) # Hide tick marks
def preprocess_img(img, skip_dilate = False):
#Kernel size: +ve, odd, square
preprocess = cv2.GaussianBlur(img.copy(), (9, 9), 0)
#cv2.adaptiveThreshold(src, maxValue, adaptiveMethod, thresholdType, blockSize, constant(c))
preprocess = cv2.adaptiveThreshold(preprocess, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, cv2.THRESH_BINARY, 11, 2)
#we need grid edges, hence,
#invert colors: gridlines will have non-zero pixels
preprocess = cv2.bitwise_not(preprocess, preprocess)
if not skip_dilate:
kernel = np.array([[0, 1, 0], [1, 1, 1], [0, 1, 0]], np.uint8)
preprocess = cv2.dilate(preprocess, kernel)
return preprocess
def find_external_contours(processed_image):
#findContours: boundaries of shapes having same intensity
new_img, ext_contours, hier = cv2.findContours(processed_image.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
new_img, contours, hier = cv2.findContours(processed_image.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
processed_image = cv2.cvtColor(processed_image, cv2.COLOR_GRAY2RGB)
#Draw all contours on image in 2px red lines
all_contours = cv2.drawContours(processed_image.copy(), contours, -1, (255, 0, 0,), 2)
external_contours = cv2.drawContours(processed_image.copy(), ext_contours, -1, (255, 0 ,0 ), 2)
#Plot images
plot_many_images([all_contours, external_contours], ['All contours', 'External Only'])
def get_corners_of_largest_poly(img):
#cv2.ContourArea(): Finds area of outermost polygon(largest feature) in img.
#Ramer Doughlas Peucker algorithm: Approximate no of sides of shape(filter rectangle objects only).
#Top Left: Smallest x and smallest y co-ordinate [minimise](x+y)
#Top Right: Largest x and smallest y co-ordinate [maximise](x-y)
#Bottom Left: Largest x and largest y co-ordinate [maximise](x+y)
#Bottom Right: Smallest x and largest y co-ordinate [minimise](x-y)
_, contours, h = cv2.findContours(img.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
contours = sorted(contours, key=cv2.contourArea, reverse=True) #Sort by area, descending
#for ele in contours:
# print(ele)
polygon = contours[0] #get largest contour
#operator.itemgetter - get index of point
bottom_right, _ = max(enumerate([pt[0][0] + pt[0][1] for pt in polygon]), key=operator.itemgetter(1))
top_left, _ = min(enumerate([pt[0][0] + pt[0][1] for pt in polygon]), key=operator.itemgetter(1))
bottom_left, _ = max(enumerate([pt[0][0] - pt[0][1] for pt in polygon]), key=operator.itemgetter(1))
top_right, _ = min(enumerate([pt[0][0] - pt[0][1] for pt in polygon]), key=operator.itemgetter(1))
print("\n"+str(bottom_right)+" "+str(bottom_left)+" "+str(top_right)+" "+str(top_left))
return [polygon[top_left][0], polygon[top_right][0], polygon[bottom_right][0], polygon[bottom_left][0]]
def display_points(in_img, points, radius=5, colour=(0, 0, 255)):
img = in_img.copy()
if len(colour) == 3:
if len(img.shape) == 2:
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
elif img.shape[2] == 1:
img = cv2.cvtColor(img, cv2.COLOR_GRAY2BGR)
for point in points:
img =, tuple(int(x) for x in point), radius, colour, -1)
return img
def display_rects(in_img, rects, colour=0):
img = in_img.copy()
for rect in rects:
img = cv2.rectangle(img, tuple(int(x) for x in rect[0]), tuple(int(x) for x in rect[1]), colour)
def infer_sudoku_puzzle(image, crop_rectangle):
#wrapPerspective: implementation of perspective transform equation.
#X = (ax + by + c) / (gx + hy + 1) X, Y -> new coords || x,y -> old coords || a..h -> constants
#Y = (dx + ey + f) / (gx + hy + 1)
#Map four coords from og img to new locations in new img
img = image
crop_rect = crop_rectangle
def distance_between(a, b): #scalar distance between a and b
#sqrt(x^2 + y^2) where (x -> ====) and (y -> ++++)
return np.sqrt( ((b[0] - a[0]) **2) + ((b[1] - a[1]) **2) )
def crop_img(): #crops rectangular portion from image and wraps it into a square of similar size
top_left, top_right, bottom_right, bottom_left = crop_rect[0], crop_rect[1], crop_rect[2], crop_rect[3]
source_rect = np.array(np.array([top_left, bottom_left, bottom_right, top_right], dtype='float32')) #float for perspective transformation
side = max([
distance_between(bottom_right, top_right),
distance_between(top_left, bottom_left),
distance_between(bottom_right, bottom_left),
distance_between(top_left, top_right)])
dest_square = np.array([[0, 0], [side - 1, 0], [side - 1, side - 1], [0, side - 1]], dtype='float32')
#Skew the image by comparing 4 before and after points -- return matrix
m = cv2.getPerspectiveTransform(source_rect, dest_square)
#Perspective Transformation on original image
return cv2.warpPerspective(img, m, (int(side), int(side)))
return crop_img()
def infer_grid(img): #infer 130 cells from image
squares = []
square_width = img.shape[0] // 10
square_height = img.shape[1] // 13
for i in range(10): #get each box and append it to squares -- 9 rows, 9 cols
for j in range(13):
p1 = (i*square_width, j*square_height + square_height*0.05) #top left corner of box
p2 = ((i+1)*square_width, (j+1)*square_height + square_height*0.05) #bottom right corner of box
squares.append((p1, p2))
kochu = int(square_height*0.05)
crop_img = img[(j*square_height + kochu):((j+1)*square_height + kochu), (i*square_width):((i+1)*square_width)]
name = "./crop_images/" + "square" + str(i) + "_" + str(j) + ".jpg"
cv2.imwrite(name, crop_img)
return squares
def main():
img = cv2.imread('002.jpg', cv2.IMREAD_GRAYSCALE)
processed_sudoku = preprocess_img(img)
corners_of_sudoku = get_corners_of_largest_poly(processed_sudoku)
display_points(processed_sudoku, corners_of_sudoku)
cropped_sudoku = infer_sudoku_puzzle(img, corners_of_sudoku)
squares_on_sudoku = infer_grid(cropped_sudoku)
display_rects(cropped_sudoku, squares_on_sudoku)
if __name__ == "__main__":
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment