CV：计算机视觉技术之图像基础知识(二)—图像内核的可视化解释

痛定思痛。 2022-12-23 11:12 102阅读 0赞

**目录**

图像内核的可视化解释

测试九种卷积核

官方Demo

DIY图片测试

DIY实时视频测试

--------------------

**相关文章**  
[CV：计算机视觉技术之图像基础知识(二)—图像内核的可视化解释][CV]  
[CV：计算机视觉技术之图像基础知识(二)—图像内核的可视化解释实现][CV 1]

# 图像内核的可视化解释 #

**原作者**： [Victor Powell][]

图像内核是一个小的矩阵，用于应用在Photoshop或Gimp中可能发现的效果，如模糊、锐化、轮廓或压纹。它们还被用于机器学习的“特征提取”，一种确定图像最重要部分的技术。在这种情况下，这个过程通常被称为“卷积”(详见[卷积神经网络][Link 1])。

为了了解它们是如何工作的，让我们从检查一个黑白图像开始。左边的矩阵包含0到255之间的数字，每个数字对应一张人脸图像中一个像素的亮度。大的、颗粒状的图像被放大，以便更容易看到;最后一个图像是“真实”大小。

![20201124185504878.gif][]

# 测试九种卷积核 #

## 官方Demo ##

*  blur
 *  bottom sobel 
 *  emboss
 *  identity 
 *  left sobel 
 *  outline
 *  right sobel 
 *  sharpen
 *  top sobel

接下来，看看如何将下面的3x3锐化内核应用到上面的一张脸的图像上。下面，对于左边图像中每3x3个像素块，我们将每个像素乘以核中对应的项，然后求和。这个和就变成了右边图像中的一个新像素。将鼠标悬停在图像上的一个像素上，看看它的值是如何计算的。

<table> 
 <tbody> 
  <tr> 
   <td style="width:48px;"><strong>blur</strong></td> 
   <td><img alt="" height="497" src="https://img-blog.csdnimg.cn/20201124185824914.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="687"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>bottom sobel&nbsp;</strong></td> 
   <td><img alt="" height="480" src="https://img-blog.csdnimg.cn/20201124185957433.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="659"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>emboss</strong></td> 
   <td><img alt="" height="468" src="https://img-blog.csdnimg.cn/20201124190145723.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="639"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>identity&nbsp;</strong></td> 
   <td><img alt="" height="468" src="https://img-blog.csdnimg.cn/20201124190223259.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="623"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>&nbsp;left sobel&nbsp;</strong></td> 
   <td><img alt="" height="461" src="https://img-blog.csdnimg.cn/20201124190312496.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="621"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>outline</strong></td> 
   <td><img alt="" height="464" src="https://img-blog.csdnimg.cn/20201124190345101.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="628"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>right sobel&nbsp;</strong></td> 
   <td><img alt="" height="462" src="https://img-blog.csdnimg.cn/20201124190443271.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="633"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>sharpen</strong></td> 
   <td><img alt="" height="470" src="https://img-blog.csdnimg.cn/20201124190604764.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="631"></td> 
  </tr> 
  <tr> 
   <td style="width:48px;"><strong>top sobel</strong></td> 
   <td><img alt="" height="456" src="https://img-blog.csdnimg.cn/20201124190653142.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70" width="621"></td> 
  </tr> 
 </tbody> 
</table>

这个过程的一个微妙之处在于如何沿着图像的边缘进行处理。例如，输入图像的左上角只有三个邻居。解决这个问题的一种方法是将原始图像的边缘值扩展一，同时保持新图像的大小不变。在这个演示中，我们将这些值设置为黑色，从而忽略了它们。

## DIY图片测试 ##

自己测试，你可以选择不同的核矩阵，看看他们如何影响原始图像或建立你自己的核。如果你的浏览器支持的话，你也可以上传你自己的图片或者使用实时视频。

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 1][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 2][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 3][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 4][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 5][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 6][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 7][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 8][]

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 9][]

## DIY实时视频测试 ##

![watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 10][]

[CV]: https://yunyaniu.blog.csdn.net/article/details/110090871
[CV 1]: https://yunyaniu.blog.csdn.net/article/details/110092162
[Victor Powell]: http://twitter.com/vicapow
[Link 1]: https://en.wikipedia.org/wiki/Convolutional_neural_network
[20201124185504878.gif]: /images/20221120/35473c20706b420dadc443e2240cb74f.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20201124191701775.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20201124191727553.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20201124191758185.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20201124191819550.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20201124191842312.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20201124191904587.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/2020112419193013.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 7]: https://img-blog.csdnimg.cn/2020112419195080.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 8]: https://img-blog.csdnimg.cn/20201124192012293.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 9]: https://img-blog.csdnimg.cn/20201124192039506.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4_size_16_color_FFFFFF_t_70 10]: https://img-blog.csdnimg.cn/20201124192341533.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3FxXzQxMTg1ODY4,size_16,color_FFFFFF,t_70