📄 chap08.htm
字号:
<html>
<!-- Mirrored from www.lgui.net/column/book1/chap08.htm by HTTrack Website Copier/3.x [XR&CO'2005], Mon, 21 Mar 2005 13:20:21 GMT -->
<head>
<meta http-equiv=Content-Type content="text/html; charset=GB2312">
<link rel=Edit-Time-Data href="chap08.files/editdata.html" >
<link rel=OLE-Object-Data href="chap08.files/oledata.mso" >
<title>第8章 图象的检测及模板匹配</title>
<style><!--
.Normal
{text-align:justify;
text-justify:inter-ideograph;
text-indent:0pt;
line-height:normal;
font-size:10.5pt;
font-family:"Times New Roman";}
.a
{text-align:center;
text-indent:0pt;
line-height:20.0pt;
font-size:12.0pt;
font-family:"Times New Roman";}
-->
</style>
</head>
<body lang=ZH-CN link=blue vlink=purple class="Normal" bgcolor="#FFFFFF">
<div style='layout-grid:15.6pt'>
<h1><a name="_Toc486331903"></a><a name="_Toc486332903"></a><a
name="_Toc486339012"></a><a name="_Toc454810877"></a><a name="_Toc454856651"><span><span>第<span
lang=EN-US>8</span></span></span></a><span><span><span style='font-family:黑体;"Times New Roman"'>章</span>
</span></span><span><span><span
style='font-family:黑体;"Times New Roman"'>图象的检测及模板匹配</span></span></span></h1>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>图象的分割与检测</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>识别</span><span lang=EN-US>)</span><span
style='font-family:宋体;"Times New Roman"'>实际上是一项非常困难的工作。很难说清楚为什么图象应该分割成这样而不是那样。人类的视觉系统是非常优越的，它不仅包含了双眼，还包括了大脑，可以从很复杂的景物中分开并识别每个物体，甚至可以毫不费力地跟上每秒好几十帧变化的图象。举两个例子来说明一下人类视觉系统的优越性。</span></p>
<table border=0 cellspacing=0 cellpadding=0>
<tr>
<td width=276 valign=top class="Normal">
<p class=a style='line-height:18.0pt'><span lang=EN-US> <img width=264 height=139
src="chap08.files/image001.gif" v:shapes="_x0000_i1025"> </span></p>
<p class=a style='line-height:18.0pt'><b><span style='font-family:宋体;
"Times New Roman"'>图</span>8.1 </b><b><span
style='font-family:宋体;"Times New Roman"'>单词</span><span lang=EN-US>THE</span></b></p>
</td>
<td width=276 valign=top class="Normal">
<p class=a style='line-height:18.0pt'><span lang=EN-US> <img width=153 height=137
src="chap08.files/image003.gif" v:shapes="_x0000_i1026"> </span></p>
<p class=a style='line-height:18.0pt'><b><span style='font-family:宋体;
"Times New Roman"'>图</span>8.2 </b><b><span
style='font-family:宋体;"Times New Roman"'>看不见的三角</span><span lang=EN-US></span></b></p>
</td>
</tr>
</table>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>图</span><span
lang=EN-US>8.1</span><span style='font-family:宋体;
"Times New Roman"'>是单词</span><span lang=EN-US>THE</span><span
style='font-family:宋体;"Times New Roman"'>，这一点很容易看出来，但仔细观察一下，就会发现，图中少了很多线条。在我们人类看来很简单的一件事，让计算机来做就很困难了。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>图</span><span
lang=EN-US>8.2</span><span style='font-family:宋体;
"Times New Roman"'>中尽管没有任何线条，但我们还是可以很容易的看出中间存在着一个白色三角形。计算机却很难发现。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>由于人类在观察图象时适用了大量的知识，所以没有任何一台计算机在分割和检测真实图象时，能达到人类视觉系统的水平。正因为如此，对于大部分图象应用来说，自动分割与检测还是一个将来时。目前只有少数的几个领域</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>如印刷体识别</span><span lang=EN-US>OCR)</span><span
style='font-family:宋体;"Times New Roman"'>自动识别达到了实用的水平。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>也许算是题外话，我们可以憧憬这样一种应用：基于内容的搜索。在一场足球比赛的录象中，用户可以输入命令，由计算机自动搜索出所有射门的镜头并显示在屏幕上。目前，我们能从一幅图象中获得的信息只是每个象素的颜色或灰度值，除此以外别无其它，完成上述功能实在是太困难了。所以说解决图象分割和检测最根本的方法是在编码</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>成象</span><span lang=EN-US>)</span><span
style='font-family:宋体;"Times New Roman"'>时就给予考虑。这也正是</span><span lang=EN-US>MPEG4</span><span
style='font-family:宋体;"Times New Roman"'>及未来的视频压缩编码标准的主要工作。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>正因为有上述的困难，所以我们今天要介绍的只是一些最基本，最简单的算法和思想，针对也只能是一些具体</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>而不是通用</span><span lang=EN-US>)</span><span
style='font-family:宋体;"Times New Roman"'>的应用。算法共有三个：投影法、差影法和模板匹配。</span></p>
<h2> <span
lang=EN-US>8.1</span> <span lang=EN-US> </span><a name="_Toc486331904"></a><a
name="_Toc486332904"></a><a name="_Toc486339013"></a><a name="_Toc454810878"></a><a
name="_Toc454856652"><span><span>投影法</span></span></a></h2>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>在介绍投影法之前，我先出一道题目，下面的这幅照片是著名的华盛顿纪念碑</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>我记得在“阿甘正传”中曾经看到过它</span><span
lang=EN-US>)</span><span style='font-family:宋体;
"Times New Roman"'>，怎样从图中自动检测到水平方向上纪念碑的位置。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>仔细观察，不难发现，纪念碑上象素的灰度都差不多而且与众不同，如果我们选取合适的阈值，做削波处理</span><span
lang=EN-US>(</span><span style='font-family:宋体;
"Times New Roman"'>这里选</span><span lang=EN-US>175</span><span
style='font-family:宋体;"Times New Roman"'>到</span><span lang=EN-US>220)</span><span style='font-family:
宋体;"Times New Roman"'>，将该图二值化，如图</span><span
lang=EN-US>8.3</span><span style='font-family:宋体;
"Times New Roman"'>所示：</span></p>
<div align=center>
<table border=0 cellspacing=0 cellpadding=0>
<tr>
<td width=276 valign=top class="Normal">
<p class=a style='line-height:18.0pt'><span lang=EN-US> <img width=216 height=216
src="chap08.files/image005.jpg" v:shapes="_x0000_i1027"> </span></p>
<p class=a style='line-height:18.0pt'><b><span style='font-family:宋体;
"Times New Roman"'>图</span>8.3 </b><b><span
style='font-family:宋体;"Times New Roman"'>华盛顿纪念碑</span><span lang=EN-US></span></b></p>
</td>
<td width=276 valign=top class="Normal">
<p class=a style='line-height:18.0pt'><span lang=EN-US> <img width=215 height=217
src="chap08.files/image007.gif" v:shapes="_x0000_i1028"> </span></p>
<p class=a style='line-height:18.0pt'><b><span style='font-family:宋体;
"Times New Roman"'>图</span>8.4 </b><b><span
style='font-family:宋体;"Times New Roman"'>削波处理，将图</span><span lang=EN-US>8.3</span></b><b><span
style='font-family:宋体;"Times New Roman"'>二值化</span><span lang=EN-US></span></b></p>
</td>
</tr>
</table>
</div>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>由于纪念碑所在的那几列的白色点比起其他列多很多，如果把该图在垂直方向做投影，如图</span><span
lang=EN-US>8.5</span><span style='font-family:宋体;
"Times New Roman"'>所示。</span></p>
<p class=a style='line-height:18.0pt'><span lang=EN-US> <img width=210 height=168
src="chap08.files/image009.jpg" v:shapes="_x0000_i1029"> </span></p>
<p align=center style='text-align:center;line-height:18.0pt'><b><span
style='font-family:宋体;"Times New Roman"'>图</span>8.5 </b><b><span
style='font-family:宋体;"Times New Roman"'>图</span><span lang=EN-US>8.4</span></b><b><span
style='font-family:宋体;"Times New Roman"'>做垂直方向投影</span><span lang=EN-US></span></b></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>其中，黑色线条的高度代表了该列上白色点的个数。图中间的高峰部分就是我们要找的水平方向上纪念碑所在的位置，这就是投影法。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>可以看出投影法是一种很自然的想法，有点象灰度直方图。为了得到更好的效果，投影法经常和阈值化一起使用。由于噪声点对投影有一定的影响，所以处理前最好先做一次平滑，去除噪声。</span></p>
<p style='line-height:18.0pt'><span style='font-family:宋体;
"Times New Roman"'>以下是投影法的源程序，第二个参数是个</span><span
lang=EN-US>BOOL</span><span style='font-family:宋体;
"Times New Roman"'>变量，为真时表示在水平方向上做投影，否则在垂直方向上做投影。要注意的是，我们针对的虽然是二值图，但为了处理的方便，用的是</span><span
lang=EN-US>256</span><span style='font-family:宋体;
"Times New Roman"'>级灰度图，不过只用到了</span><span lang=EN-US>0</span><span
style='font-family:宋体;"Times New Roman"'>和</span><span lang=EN-US>255</span><span style='font-family:
宋体;"Times New Roman"'>两种灰度级。</span></p>
<p style='line-height:18.0pt'><span lang=EN-US>BOOL Projection(HWND hWnd,BOOL
Hori)</span></p>
<p style='line-height:18.0pt'><span lang=EN-US>{</span></p>
<p style='line-height:18.0pt'><span> </span>DWORD
OffBits,BufSize;</p>
<p style='line-height:18.0pt'>LPBITMAPINFOHEADER lpImgData;</p>
<p style='line-height:18.0pt'><span> </span>LPSTR
lpPtr;</p>
<p style='line-height:18.0pt'><span> </span>HLOCAL
hTempImgData;</p>
<p style='line-height:18.0pt'><span> </span>LPBITMAPINFOHEADER
lpTempImgData;</p>
<p style='line-height:18.0pt'><span> </span>LPSTR
⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -