铅笔素描算法

by Rishav Agarwal

通过里沙夫·阿加瓦尔

如何用10行代码将任何图像变成铅笔素描 (How to turn any image into a pencil sketch with 10 lines of code)

使用基本的计算机视觉和Python的Numpy库 (Use basic computer vision and Python’s Numpy library)

I have always been fascinated by computer vision, and especially by its power to manipulate images.

我一直着迷于计算机视觉，尤其是其操纵图像的能力。

An image is basically an array of numbers to Python. So we can perform a variety of matrix manipulations to get some very interesting results. In this post, I talk about how to reduce an image into a ‘pencil’ outline.

图像基本上是Python的数字数组。 因此，我们可以执行各种矩阵操作来获得一些非常有趣的结果。 在本文中，我将讨论如何将图像缩小为“铅笔”轮廓。

步骤 (The Steps)

The process is pretty simple:

这个过程非常简单：

1. Grayscale the image灰度图像
2. Invert it倒转
3. Blur the inverted image模糊倒置图像
4. Dodge blend the blurred and grayscale image.闪避将模糊的灰度图像混合。

We can pick any image from the Internet. I will go with this image of Indian cricketer Virat Kohli:

我们可以从互联网上选择任何图像。 我将使用印度板球运动员Virat Kohli的这张照片：

1.加载图片 (1. Load image)

import imageioimg="http://static.cricinfo.com/db/PICTURES/CMS/263600/263697.20.jpg"start_img = imageio.imread(img)

You can see how Python sees this image with the shape attribute:

您可以看到Python如何看到带有shape属性的图像：

start_img.shape(196, 160, 30)

So this is a three channel image of size 196x160.

因此，这是一个大小为196x160的三通道图像。

2.灰度 (2. Greyscale)

We then make the image black and white.

然后，我们将图像设为黑白。

Numpy doesn’t have any in-built function for grayscaling, but we can easily convert the image using the formula. You can learn why this formula works right here.

Numpy没有任何内置的灰度功能，但是我们可以使用公式轻松转换图像。 您可以在这里了解为什么使用此公式。

Y= 0.299 R + 0.587 G + 0.114 B

So our function will look like:

因此我们的函数将如下所示：

import numpy as npdef grayscale(rgb): return np.dot(rgb[...,:3], [0.299, 0.587, 0.114])

Applying grayscale:

应用灰度：

gray_img = grayscale(start_img)

3.反转图像 (3. Invert image)

We can invert images simply by subtracting from 255, as grayscale images are 8 bit images or have a maximum of 256 tones.

我们可以简单地通过减去255来反转图像，因为灰度图像是8位图像或最多具有256个色调。

inverted_img = 255-gray_img

4.图像模糊 (4. Blur Image)

We now blur the inverted image. Blurring is done by applying a Gaussian filter to the inverted image. The key here is the variance of the Gaussian function or sigma.

现在，我们对倒置的图像进行模糊处理。 通过对反转图像应用高斯滤波器来完成模糊处理。 此处的关键是高斯函数或sigma的方差。

As sigma increases, the image becomes more blurred. Sigma controls the extent of the variance and thus, the degree of blurring.

随着sigma的增加，图像变得更加模糊。 Sigma控制差异的程度，从而控制模糊程度。

import scipy.ndimageblur_img = scipy.ndimage.filters.gaussian_filter(inverted_img,sigma=5)

5.闪避和合并 (5. Dodge and Merge)

The Colour Dodge blend mode divides the bottom layer by the inverted top layer. This lightens the bottom layer depending on the value of the top layer. We have the blurred image, which highlights the boldest edges.

“ 颜色减淡”混合模式将底层除以反转的顶层。 这会根据顶层的值减轻底层的重量。 我们有模糊的图像，该图像突出显示了最大胆的边缘。

As all our images are read using Numpy, all the matrix calculations are super fast.

由于我们使用Numpy读取所有图像，因此所有矩阵计算都非常快。

def dodge(front,back): result=front*255/(255-back)  result[result>255]=255 result[back==255]=255 return result.astype(‘uint8’)

final_img= dodge(blur_img,gray_img)

And that’s it!

就是这样！

6.绘制并保存 (6. Plot and save)

We can plot our final image using plt.imgshow. Note that we need to keep the cmap argument equal to “gray”.

我们可以使用plt.imgshow绘制最终图像。 注意，我们需要保持cmap参数等于“gray” 。

import matplotlib.pyplot as pltplt.imshow(final_img, cmap=”gray”)

We can save the image using:

我们可以使用以下方法保存图像：

plt.imsave(‘img2.png’, final_img, cmap=’gray’, vmin=0, vmax=255)

最后结果 (Final result)

整个代码生效 (Entire code in action)

Here we don’t have much room to play with, except with the sigma parameter while blurring.

在这里，除了模糊时使用sigma参数，我们没有太多玩的余地。

As sigma increases, the pic becomes clearer but run time also increases. So a sigma of 5 works well for us.

随着sigma的增加，图片变得更清晰，但运行时间也会增加。 因此5的sigma对我们来说效果很好。

简码： (Condensed code:)

I promised 10 lines or less, so here you go:

我答应不超过10行，因此您可以：

As always, you can find the entire detailed code on my GitHub.

与往常一样，您可以在我的GitHub上找到完整的详细代码。

PS this is how I created my Medium DP. If you like this blog, show some ❤️ :)

PS这就是我创建中型DP的方式。 如果您喜欢此博客，请显示一些❤️:)

Also I don’t own this image of Virat. I hope he doesn’t mind!

我也不拥有Virat的形象。 我希望他不介意！

翻译自: https://www.freecodecamp.org/news/sketchify-turn-any-image-into-a-pencil-sketch-with-10-lines-of-code-cf67fa4f68ce/

铅笔素描算法