How to construct horizontal projection of binary image in OpenCV

问题

I am doing a text segmentation project for school. I need to do horizontal image projection of a binary image. The results that I want are like this:

I am using OpenCV in Python. I used x_sum = cv2.reduce(img, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S) to get the array of sums, as advised by this question: horizontal and vertical projection of an image and this question: Horizontal Histogram in OpenCV.

I tried to get the horizontal projection image by using cv2.calcHist, but what I got was just a single horizontal line. My code is below:

image = cv2.imread(file_name)
x_sum = cv2.reduce(image, 0, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
horizontal_projection=cv2.calcHist(x_sum,[0],None,[256],[0,256])
cv2.imwrite("image2.png", horizontal_projection)

Please help and tell me what I am doing wrong. I need my horizontal projection results to be just like the Quora example.

回答1:

When calculating the projection, you basically want to sum the pixels along each row of the image. However, your text is black, which is encoded as zero so you will get small numbers where there is a lot of text in a row and large numbers where there is little text in a row - which is the opposite of what you want - so you need to invert:

import cv2
import numpy as np

# Load as greyscale
im = cv2.imread('text.png', cv2.IMREAD_GRAYSCALE)

# Invert
im = 255 - im

# Calculate horizontal projection
proj = np.sum(im,1)

The array proj is now 141 rows tall, each corresponding to how much text is in that row of the image:

array([    0,     0,     0,     0,    40,    44,   144,   182,   264,
         326,   425,  1193,  2718,  5396,  9272, 11880, 13266, 13597,
       12906, 11962, 10791,  9647,  8554, 20469, 45426, 65714, 81397,
       81675, 66590, 58714, 58046, 60516, 66136, 71794, 77552, 78555,
       74868, 72083, 70139, 70160, 72174, 76409, 82854, 88962, 94721,
       88105, 69126, 47753, 23966, 13845, 17406, 19145, 19079, 16548,
       11524,  8511,  7465,  7042,  7197,  6577,  5022,  3476,  1797,
         809,   450,   309,   348,   351,   250,   232,   271,   279,
         251,   628,  1419,  3259,  6187,  8272,  9551,  9825,  9119,
        7984,  6444,  5305,  4596, 13385, 31647, 46330, 57459, 56139,
       42402, 34928, 33729, 35055, 38874, 41649, 43394, 43265, 41291,
       40126, 39767, 40515, 42390, 44478, 46793, 47881, 47743, 43983,
       36644, 28054, 18242, 15583, 20047, 22038, 21569, 17751, 10571,
        6830,  6580,  6231,  5681,  4595,  2879,  1642,   771,   365,
         320,   282,   105,    88,    76,    76,    28,    28,    28,
          28,     0,     0,     0,     0,     0], dtype=uint64)

I cropped your image to 819x141 pixels as follows:

There are many ways to do the visualisation. Here is one:

#!/usr/bin/env python3

import cv2
import numpy as np

# Load as greyscale
im = cv2.imread('text.png', cv2.IMREAD_GRAYSCALE)

# Invert
im = 255 - im

# Calculate horizontal projection
proj = np.sum(im,1)

# Create output image same height as text, 500 px wide
m = np.max(proj)
w = 500
result = np.zeros((proj.shape[0],500))

# Draw a line for each row
for row in range(im.shape[0]):
   cv2.line(result, (0,row), (int(proj[row]*w/m),row), (255,255,255), 1)

# Save result
cv2.imwrite('result.png', result)

回答2:

Try This Simple Code

 import cv2
 import numpy as np
 from matplotlib import pyplot as plt
 img = cv2.imread('text.png', cv2.IMREAD_GRAYSCALE)
 height, width = img.shape[:2]
 med = cv2.medianBlur(img,3)
 cv2.namedWindow("med", cv2.WINDOW_NORMAL)
 cv2.imshow("med",med)
 cv2.waitKey(0)
 sum_x = cv2.reduce(cv2.bitwise_not(med), 1, cv2.REDUCE_SUM, dtype=cv2.CV_32S)
 plt.plot(sum_x)
 plt.xlim([0, height])
 plt.show()

来源：https://stackoverflow.com/questions/54285839/how-to-construct-horizontal-projection-of-binary-image-in-opencv

标签

python

OpenCV

image-processing

image-segmentation