根据暗中心线使用python水平分割图像 [英] Split image horizontally with python based on dark center line

问题描述

目标是基于比周围其他区域更暗的垂直中心线，在python中水平分割图像(双报纸页面).

Goal is to horizontally split an image (double newspaper page) in python based on a vertical centerline that is darker than other areas around.

示例图片:

使用opencv(cv2)进行初始裁剪时有些运气，并使用cv2.Canny从黑色背景旋转双页，然后根据cv2.contourArea对轮廓进行排序.

Had some luck using opencv (cv2) for the initial crop and rotation of the double page from a black background using cv2.Canny, and then sorting the contours based on cv2.contourArea.

但是，现在我只是想找到一条中心线，然后将图像分成两个单独的图像.再次使用cv2.Canny，我看到它可以识别该中心线，但是不确定如何识别该长的垂直线并将其用于分割图像:

But, now I'm just interested in finding a center line and then splitting the image into two separate images. Using cv2.Canny again I see that it's able to identify that centerline, but not sure how to identify that long, vertical line and use that to split the image:

最终目标将是两个图像，如下所示:

End goal would be two images like the following:

任何建议都将受到欢迎.

Any suggestions would be most welcome.

推荐答案

首先，运行水平渐变，以便仅突出垂直边缘.您可以使用以下系数计算水平梯度:

First, run a horizontal gradient so you only accentuate vertical edges. You can calculate a horizontal gradient with these coefficients:

-1   0   1
-2   0   2
-1   0   1

然后计算垂直列的总和，您可以使用np.sum(array,axis=0)，您将获得以下信息:

Then compute the sum of the vertical columns, you can use np.sum(array,axis=0) and you will get this:

为了便于查看，我对其进行了重新整形-实际上它只有1像素高.希望您可以在中间看到Numpy argmax()处的亮白线.当您只进行水平渐变时，效果也会更好，因为此刻我正在使用紫色和黄色图像，并增强了垂直和水平边缘.

I have re-shaped it for ease of viewing - it is actually only 1 pixel tall. Hopefully you can see the bright white line in the middle which you can find with Numpy argmax(). It will also be better when you just do a horizontal gradient because at the moment I am using the purple and yellow image with vertical and horizontal edges enhanced.

请注意，此方法的灵感在于您说"要识别出较长的垂直中心线" ，其理由是一长行白色像素将加起来很大.请注意，我已经假设您的图像是歪斜的(因为您说这条线是垂直的)，并且此方法在"vertical" 线会分布在几列的歪斜图像上可能效果不佳.

Note that the inspiration for this approach is that you said you "want to identify that long, vertical centerline" and the rationale is that a long line of white pixels will add up to a large sum. Note that I have assumed your image is de-skewed (since you said the line is vertical) and this method may not work so well on skew images where "vertical" line will be spread across several columns.

这篇关于根据暗中心线使用python水平分割图像的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！