a.rar_region growing_论文 word_图形图像处理

<html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta charset="utf-8"> <meta name="generator" content="pdf2htmlEX"> <meta http-equiv="X-UA-Compatible" content="IE=edge,chrome=1"> <link rel="stylesheet" href="https://static.pudn.com/base/css/base.min.css"> <link rel="stylesheet" href="https://static.pudn.com/base/css/fancy.min.css"> <link rel="stylesheet" href="https://static.pudn.com/prod/directory_preview_static/6253973c6caf596192a391d7/raw.css"> <script src="https://static.pudn.com/base/js/compatibility.min.js"></script> <script src="https://static.pudn.com/base/js/pdf2htmlEX.min.js"></script> <script> try{ pdf2htmlEX.defaultViewer = new pdf2htmlEX.Viewer({}); }catch(e){} </script> <title></title> </head> <body> <div id="sidebar" style="display: none"> <div id="outline"> </div> </div> <div id="pf1" class="pf w0 h0" data-page-no="1"><div class="pc pc1 w0 h0"><img class="bi x0 y0 w1 h1" alt="" src="https://static.pudn.com/prod/directory_preview_static/6253973c6caf596192a391d7/bg1.jpg"><div class="t m0 x1 h2 y1 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x2 h3 y2 ff2 fs0 fc0 sc0 ls0 ws0">—<span class="ff1">217</span>—</div><div class="t m0 x3 h4 y3 ff3 fs1 fc0 sc0 ls1 ws0"> </div><div class="t m0 x3 h4 y4 ff3 fs1 fc0 sc0 ls1 ws0"> </div><div class="t m0 x3 h4 y5 ff3 fs1 fc0 sc0 ls1 ws0"> <span class="fs2 ls0"> </span></div><div class="t m0 x4 h5 y6 ff4 fs3 fc0 sc1 ls2 ws0">基于区域生长的鱼眼图像轮廓提取算法</div><div class="t m0 x5 h6 y7 ff3 fs3 fc0 sc0 ls0 ws0"> </div><div class="t m0 x6 h7 y8 ff5 fs0 fc0 sc1 ls3 ws0">杨 <span class="_ _0"> </span>丹</div><div class="t m0 x7 h8 y9 ff3 fs4 fc0 sc0 ls0 ws0">1</div><div class="t m0 x8 h7 y8 ff5 fs0 fc0 sc1 ls3 ws0">，游 <span class="_ _0"> </span>磊</div><div class="t m0 x9 h8 y9 ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 xa h7 y8 ff5 fs0 fc0 sc1 ls3 ws0">，张小洪</div><div class="t m0 xb h8 y9 ff3 fs4 fc0 sc0 ls0 ws0">1</div><div class="t m0 xc h7 y8 ff5 fs0 fc0 sc1 ls3 ws0">，李 <span class="_ _0"> </span>博</div><div class="t m0 xd h8 y9 ff3 fs4 fc0 sc0 ls0 ws0">3</div><div class="t m0 xe h9 y8 ff3 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 xf ha ya ff1 fs5 fc0 sc0 ls4 ws0">(1. <span class="_"> </span><span class="ff5 ls0">重庆大学软件学院，重庆</span><span class="ls5 ws1"> 400030</span><span class="ff5 ls0">；</span><span class="ls5">2. <span class="_"> </span><span class="ff5 ls0">重庆大学数理学院，重庆</span><span class="ws1"> 400030</span><span class="ff5 ls0">；</span><span class="ls6"> </span></span></div><div class="t m0 x10 ha yb ff1 fs5 fc0 sc0 ls5 ws0">3. <span class="_"> </span><span class="ff5 ls0">重庆大学计算机学院，重庆</span><span class="ws1"> 400030)</span></div><div class="t m0 x11 hb yc ff1 fs6 fc0 sc0 ls0 ws0"> </div><div class="t m0 x3 ha yd ff5 fs5 fc0 sc1 ls0 ws0">摘<span class="ff3 sc0 ls6"> </span><span class="ls7">要：</span><span class="sc0">以图像二值化为前提，使用基于区域生长的改进算法，获取图像轮廓，<span class="_ _1"></span>实现鱼眼图像校正，克服鱼眼图像轮廓区域内出现大量黑</span></div><div class="t m0 x3 ha ye ff5 fs5 fc0 sc0 ls0 ws0">色像素使面积统计算法失效的缺点。仿真实验表明，该算法精度高、效率高，能够获得满意的校正结果。<span class="ff1"> </span></div><div class="t m0 x3 ha yf ff5 fs5 fc0 sc1 ls7 ws0">关键词：<span class="sc0 ls0">鱼眼图像；区域生长；轮廓提取；非线性最小二乘拟合<span class="ff1"> </span></span></div><div class="t m0 x12 hc y10 ff3 fs7 fc0 sc0 ls8 ws2">Fisheye <span class="_ _2"></span>Image <span class="_ _2"></span>Contour<span class="_ _1"></span> <span class="_ _2"></span>Extraction <span class="_ _3"></span>Algorithm </div><div class="t m0 x13 hc y11 ff3 fs7 fc0 sc0 ls9 ws3">Based on Region Gr<span class="_ _1"></span>owing</div><div class="t m0 x14 hd y12 ff6 fs8 fc0 sc1 ls0 ws0"> </div><div class="t m0 x15 h9 y13 ff3 fs0 fc0 sc0 lsa ws4">Y<span class="_ _4"></span>ANG<span class="_ _1"></span> Dan</div><div class="t m0 x16 h8 y14 ff3 fs4 fc0 sc0 ls0 ws0">1</div><div class="t m0 x17 h9 y13 ff3 fs0 fc0 sc0 lsb ws5">, YOU <span class="_ _5"></span>Lei</div><div class="t m0 x18 h8 y14 ff3 fs4 fc0 sc0 ls0 ws0">2</div><div class="t m0 x19 h9 y13 ff3 fs0 fc0 sc0 lsc ws6">, ZHANG<span class="_ _1"></span> Xiao-hong</div><div class="t m0 x1a h8 y14 ff3 fs4 fc0 sc0 ls0 ws0">1</div><div class="t m0 x1b h9 y13 ff3 fs0 fc0 sc0 lsd ws7">, LI Bo</div><div class="t m0 x1c h8 y14 ff3 fs4 fc0 sc0 ls0 ws0">3</div><div class="t m0 x1d h9 y13 ff3 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m0 x1e he y15 ff1 fs5 fc0 sc0 lse ws8">(1. College of Software, Chongqing University<span class="_ _4"></span>, Chongqing 400030; 2. Co<span class="lsf ws9">llege of Mathematics &<span class="_ _1"></span> Physics, Chongqing University<span class="_ _1"></span>,<span class="_ _1"></span> <span class="_ _0"> </span> <span class="_"> </span> <span class="_"> </span> <span class="_"> </span><span class="ls6 ws0"> </span></span></div><div class="t m0 x1f he y16 ff1 fs5 fc0 sc0 ls10 wsa">Chongqing 400030; 3. College of Comput<span class="_ _1"></span><span class="ls11 wsb">er,<span class="_ _1"></span> Chongqing University<span class="_ _1"></span>, Chongqing 400030)</span></div><div class="t m0 x20 hb y17 ff1 fs6 fc0 sc0 ls0 ws0"> </div><div class="t m0 x3 ha y18 ff5 fs5 fc0 sc1 ls0 ws0">【<span class="ff3 sc0 ls12">Abstract</span>】<span class="ff1 sc0 ls13 wsc">On the premise of image binary<span class="_ _4"></span>, image contour<span class="_"> </span>is obtained by <span class="ls14 wsd">m<span class="_ _1"></span>eans of an improved algorithm <span class="ls10 wse">based on re<span class="_ _1"></span>gion growing, image</span></span></span></div><div class="t m0 x3 he y19 ff1 fs5 fc0 sc0 ls15 wsf">correction can be completed. It can overcome the shortage<span class="_"> </span>that ma<span class="ls16 ws10">ssive black pixels appear in th<span class="ls17 ws11">e fisheye image circular outline<span class="ls18 ws12"> area and make the</span></span></span></div><div class="t m0 x3 he y1a ff1 fs5 fc0 sc0 ls19 ws13">statistical algorithm fail. Simulation results show that the proposed algorithm has advantage <span class="lse ws14">of high precision and high ef<span class="_ _1"></span>fici<span class="ls12 ws15">ency<span class="_ _1"></span>, satisfactory results</span></span></div><div class="t m0 x3 he y1b ff1 fs5 fc0 sc0 ls1a ws16">are obtained. </div><div class="t m0 x3 ha y1c ff5 fs5 fc0 sc1 ls0 ws0">【<span class="ff3 sc0 ls1b ws17">Key wo<span class="_ _5"></span>rds</span>】<span class="ff1 sc0 ls15 ws18">fisheye image; region growing; contour extraction; nonlinear least square fitting</span><span class="ff3 sc0"> </span></div><div class="t m0 x21 hf y1d ff6 fs1 fc0 sc1 ls0 ws0">计<span class="ff3 sc0 ls1"> </span>算<span class="ff3 sc0 ls1"> </span>机<span class="ff3 sc0 ls1"> </span>工<span class="ff3 sc0 ls1"> </span>程</div><div class="t m0 x16 h4 y1e ff3 fs1 fc0 sc0 ls1c ws19">Computer Engineering</div><div class="t m0 x22 hf y1f ff6 fs1 fc0 sc1 ls0 ws0">第<span class="_ _6"> </span><span class="ff3 sc0 ls1d">36<span class="_"> </span></span>卷<span class="ff3 sc0 ls1"> </span>第<span class="_ _6"> </span><span class="ff3 sc0">8<span class="_"> </span></span>期<span class="ff3 sc0"> </span></div><div class="t m0 x22 h10 y20 ff7 fs1 fc0 sc0 ls1e ws1a">V<span class="_ _7"></span>ol.36 No. <span class="_ _8"></span>8<span class="ff1 ls0 ws0"> </span></div><div class="t m0 x23 hf y21 ff3 fs1 fc0 sc0 ls1d ws0">2010<span class="_"> </span><span class="ff6 sc1 ls0">年<span class="_ _6"> </span></span><span class="ls0">4<span class="_"> </span><span class="ff6 sc1">月</span></span></div><div class="t m0 x24 h4 y22 ff3 fs1 fc0 sc0 ls1 ws0"> A</div><div class="c x25 y23 w2 h11"><div class="t m0 x26 h4 y24 ff3 fs1 fc0 sc0 ls0 ws0">p</div></div><div class="t m0 x27 h4 y25 ff3 fs1 fc0 sc0 ls1f ws1b">ril 2010</div><div class="t m0 x28 hf y26 ff6 fs1 fc0 sc1 ls7 ws0">·图形图像处理·<span class="ff1 sc0 ls0"> </span></div><div class="t m0 x29 h12 y27 ff6 fs5 fc0 sc1 ls7 ws0">文章编号：<span class="ff3 sc0 ls5">1000</span><span class="ff5 ls0">—<span class="ff3 sc0 ls5">3428(2010)<span class="_ _1"></span>08<span class="ff5 sc1 ls0">—</span>0217<span class="ff5 sc1 ls0">—</span><span class="ls20">02</span></span></span></div><div class="t m0 x2a h12 y28 ff6 fs5 fc0 sc1 ls7 ws0">文献标识码：<span class="ff3 sc0 ls21 ws1c">A <span class="ff1 ls0 ws0"> </span></span></div><div class="t m0 x2b h12 y29 ff6 fs5 fc0 sc1 ls7 ws0">中图分类号：<span class="ff3 sc0 ls22">TP391.41</span></div><div class="t m1 x2c h4 y2a ff3 fs1 fc0 sc0 ls23 ws1d">1 </div><div class="t m2 x2d h13 y2a ff5 fs1 fc0 sc1 ls24 ws0">概述</div><div class="t m1 x2e h4 y2a ff3 fs1 fc0 sc0 ls0 ws0"> </div><div class="t m3 x1 h7 y2b ff5 fs0 fc0 sc0 ls25 ws0">在自动驾驶、<span class="_ _1"></span>虚拟现实等许多计算机视觉领<span class="_ _1"></span>域，需要获</div><div class="t m3 x2c h7 y2c ff5 fs0 fc0 sc0 ls26 ws0">得大尺度、<span class="_ _4"></span>广视角的全景图像。<span class="_ _7"></span>因此，<span class="_ _4"></span>全景图像拼接技术</div><div class="t m4 x2f h14 y2d ff1 fs4 fc0 sc0 ls27 ws0">[1-2]</div><div class="t m3 x2c h7 y2e ff5 fs0 fc0 sc0 ls25 ws0">成为计算机图像图形学<span class="_ _1"></span>和计算机视觉的研究热点。<span class="_ _1"></span>如果使用</div><div class="t m3 x2c h7 y2f ff5 fs0 fc0 sc0 ls26 ws0">普通镜头，需要拍<span class="_ _5"></span>摄多幅图像，经过拼接，才能获<span class="_ _5"></span>得建立在</div><div class="t m3 x2c h7 y30 ff5 fs0 fc0 sc0 ls26 ws0">柱面或者球面模型上的全景图像。<span class="_ _2"></span>这是一项耗时巨大的工作。</div><div class="t m3 x2c h7 y31 ff5 fs0 fc0 sc0 ls26 ws0">如果采用鱼眼镜头摄像机获取的图像可以包含</div><div class="t m5 x30 h2 y32 ff1 fs0 fc0 sc0 ls28 ws0">180º</div><div class="t m3 x8 h7 y32 ff5 fs0 fc0 sc0 ls26 ws0">的信息，</div><div class="t m3 x2c h7 y33 ff5 fs0 fc0 sc0 ls26 ws0">那么只需要</div><div class="t m5 x31 h2 y34 ff1 fs0 fc0 sc0 ls0 ws0">2</div><div class="t m3 x32 h7 y34 ff5 fs0 fc0 sc0 ls26 ws0">张鱼眼图像就可以包含</div><div class="t m5 x33 h2 y34 ff1 fs0 fc0 sc0 ls28 ws0">360º</div><div class="t m3 x34 h7 y34 ff5 fs0 fc0 sc0 ls26 ws0">的所有信息，即生</div><div class="t m3 x2c h7 y35 ff5 fs0 fc0 sc0 ls0 ws0">成</div><div class="t m5 x35 h2 y36 ff1 fs0 fc0 sc0 ls29 ws0">360º</div><div class="t m3 x36 h7 y36 ff5 fs0 fc0 sc0 ls26 ws0">全景图</div><div class="t m4 x4 h14 y37 ff1 fs4 fc0 sc0 ls27 ws0">[2]</div><div class="t m3 x37 h7 y36 ff5 fs0 fc0 sc0 ls26 ws0">。虽然鱼眼镜头视角较大，可以拍摄到前方</div><div class="t m3 x2c h7 y38 ff5 fs0 fc0 sc0 ls26 ws0">半球面空间的大量信息</div><div class="t m4 x38 h14 y39 ff1 fs4 fc0 sc0 ls27 ws0">[3]</div><div class="t m3 x39 h7 y3a ff5 fs0 fc0 sc0 ls26 ws0">，但是鱼眼图像有较大的失真</div><div class="t m4 x3a h14 y39 ff1 fs4 fc0 sc0 ls27 ws0">[4-5<span class="_ _1"></span>]</div><div class="t m3 x3b h7 y3a ff5 fs0 fc0 sc0 ls0 ws0">，</div><div class="t m3 x2c h7 y3b ff5 fs0 fc0 sc0 ls26 ws0">除了图像中心的景<span class="_ _5"></span>物外，其他景物会发生一定变形<span class="_ _5"></span>，不符合</div><div class="t m3 x2c h7 y3c ff5 fs0 fc0 sc0 ls26 ws0">人类的正常视觉习<span class="_ _5"></span>惯，因此，产生了从鱼眼图像恢<span class="_ _5"></span>复实际场</div><div class="t m3 x2c h7 y3d ff5 fs0 fc0 sc0 ls25 ws0">景的问题，即鱼眼图像<span class="_ _1"></span>校正。从理论上，鱼<span class="_ _1"></span>眼图像可以完全</div><div class="t m3 x2c h7 y3e ff5 fs0 fc0 sc0 ls25 ws0">恢复出真实场景<span class="_ _1"></span>。根据鱼眼摄像机的工作原<span class="_ _1"></span>理可知，鱼眼图</div><div class="t m3 x2c h7 y3f ff5 fs0 fc0 sc0 ls2a ws0">像校正</div><div class="t m4 x3c h14 y40 ff1 fs4 fc0 sc0 ls27 ws0">[6]</div><div class="t m3 x1e h7 y41 ff5 fs0 fc0 sc0 ls2a ws0">的前提条件是找到光学中心</div><div class="t m5 x33 h2 y41 ff1 fs0 fc0 sc0 ls0 ws0">(</div><div class="t m6 x3d h15 y41 ff8 fs0 fc0 sc0 ls0 ws0">x</div><div class="t m4 x3e h14 y42 ff1 fs4 fc0 sc0 ls0 ws0">0</div><div class="t m5 x3f h2 y41 ff1 fs0 fc0 sc0 ls0 ws0">,</div><div class="t m6 x40 h15 y41 ff8 fs0 fc0 sc0 ls0 ws0">y</div><div class="t m4 x41 h14 y42 ff1 fs4 fc0 sc0 ls0 ws0">0</div><div class="t m5 x42 h2 y41 ff1 fs0 fc0 sc0 ls0 ws0">)</div><div class="t m3 x43 h7 y41 ff5 fs0 fc0 sc0 ls2a ws0">以及鱼眼图像半</div><div class="t m3 x2c h7 y43 ff5 fs0 fc0 sc0 ls0 ws0">径</div><div class="t m6 x44 h15 y43 ff8 fs0 fc0 sc0 ls0 ws0">r</div><div class="t m3 x45 h7 y43 ff5 fs0 fc0 sc0 ls2b ws0">，即提取鱼眼图像轮廓。</div><div class="t m5 x46 h2 y43 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m3 x1 h7 y44 ff5 fs0 fc0 sc0 ls26 ws0">提取图像轮廓</div><div class="t m4 x47 h14 y45 ff1 fs4 fc0 sc0 ls27 ws0">[7]</div><div class="t m3 x48 h7 y44 ff5 fs0 fc0 sc0 ls26 ws0">目前主要有</div><div class="t m5 x49 h2 y44 ff1 fs0 fc0 sc0 ls0 ws0">2</div><div class="t m3 x4a h7 y44 ff5 fs0 fc0 sc0 ls26 ws0">种算法：</div><div class="t m5 x4b h2 y44 ff1 fs0 fc0 sc0 ls2c ws0">(1)</div><div class="t m3 x4c h7 y44 ff5 fs0 fc0 sc0 ls26 ws0">面积统计法</div><div class="t m4 x4d h14 y45 ff1 fs4 fc0 sc0 ls27 ws0">[8]</div><div class="t m3 x18 h7 y44 ff5 fs0 fc0 sc0 ls0 ws0">，</div><div class="t m3 x2c h7 y46 ff5 fs0 fc0 sc0 ls25 ws0">该方法速度快、<span class="_ _1"></span>实现容易，但在图像中出现<span class="_ _1"></span>较多的黑色像素</div><div class="t m3 x2c h7 y47 ff5 fs0 fc0 sc0 ls26 ws0">点时，<span class="_ _7"></span>将会产生较大误差，<span class="_ _7"></span>不再适用<span class="_ _1"></span>。</div><div class="t m5 x4e h2 y48 ff1 fs0 fc0 sc0 ls2d ws0">(2)</div><div class="t m3 x40 h7 y48 ff5 fs0 fc0 sc0 ls26 ws0">扫描线逼近算法</div><div class="t m4 x4d h14 y49 ff1 fs4 fc0 sc0 ls27 ws0">[9]</div><div class="t m3 x18 h7 y48 ff5 fs0 fc0 sc0 ls0 ws0">，</div><div class="t m3 x2c h7 y4a ff5 fs0 fc0 sc0 ls26 ws0">同样受到黑色像素<span class="_ _5"></span>点的影响，如果大量黑色像素点<span class="_ _5"></span>在图像的</div><div class="t m3 x2c h7 y4b ff5 fs0 fc0 sc0 ls26 ws0">边界处，<span class="_ _9"></span>就会产生误差。<span class="_ _9"></span>如果鱼眼图像只有一部分在图像中，</div><div class="t m3 x2c h7 y4c ff5 fs0 fc0 sc0 ls2b ws0">则上述</div><div class="t m5 x4f h2 y4c ff1 fs0 fc0 sc0 ls0 ws0">2</div><div class="t m3 x1e h7 y4c ff5 fs0 fc0 sc0 ls2b ws0">种算法将会产生更大误差。</div><div class="t m5 x50 h2 y4c ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m3 x1 h7 y4d ff5 fs0 fc0 sc0 ls26 ws0">本文介绍鱼眼图像的成像原理</div><div class="t m4 x51 h14 y4e ff1 fs4 fc0 sc0 ls27 ws0">[10]</div><div class="t m3 x52 h7 y4d ff5 fs0 fc0 sc0 ls26 ws0">，<span class="_ _4"></span>提出一种基于区域生</div><div class="t m3 x2c h7 y4f ff5 fs0 fc0 sc0 ls26 ws0">长的改进校正算法，克服了以上缺点，<span class="_ _1"></span>利用阈值分割</div><div class="t m4 x53 h14 y50 ff1 fs4 fc0 sc0 ls27 ws0">[11]</div><div class="t m3 x54 h7 y51 ff5 fs0 fc0 sc0 ls2e ws0">，将</div><div class="t m3 x2c h7 y52 ff5 fs0 fc0 sc0 ls26 ws0">鱼眼图像二值化，<span class="_ _5"></span>通过研究有效区域像素的领域特<span class="_ _5"></span>征标记二</div><div class="t m3 x2c h7 y53 ff5 fs0 fc0 sc0 ls26 ws0">值图像中黑色像素<span class="_ _5"></span>点，区分黑色像素中的有效区域<span class="_ _5"></span>与无效区</div><div class="t m3 x2c h7 y54 ff5 fs0 fc0 sc0 ls26 ws0">域，快速地获取正<span class="_ _5"></span>确的轮廓区域，根据鱼眼图像成<span class="_ _5"></span>像原理恢</div><div class="t m3 x55 h7 y55 ff5 fs0 fc0 sc0 ls2b ws0">复出实际全景图。</div><div class="t m5 x56 h2 y56 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m1 x55 h4 y57 ff3 fs1 fc0 sc0 ls23 ws1d">2 </div><div class="t m2 x57 h13 y57 ff5 fs1 fc0 sc1 ls24 ws0">鱼眼图像成像原理</div><div class="t m1 x58 h4 y57 ff3 fs1 fc0 sc0 ls0 ws0"> </div><div class="t m3 x59 h7 y2c ff5 fs0 fc0 sc0 ls26 ws0">鱼眼镜头具有</div><div class="t m5 x5a h2 y2c ff1 fs0 fc0 sc0 ls29 ws0">180º</div><div class="t m3 x5b h7 y2c ff5 fs0 fc0 sc0 ls26 ws0">的视角，可以看成一个半球，并且满</div><div class="t m3 x55 h7 y2e ff5 fs0 fc0 sc0 ls26 ws0">足球面投影模型。<span class="_ _1"></span>该投影方式可以分为</div><div class="t m5 x5c h2 y2e ff1 fs0 fc0 sc0 ls0 ws0">2</div><div class="t m3 x5d h7 y2e ff5 fs0 fc0 sc0 ls26 ws0">个过程：</div><div class="t m5 x5e h2 y2e ff1 fs0 fc0 sc0 ls2c ws0">(1)</div><div class="t m3 x5f h7 y2e ff5 fs0 fc0 sc0 ls26 ws0">每一个</div><div class="t m3 x55 h7 y2f ff5 fs0 fc0 sc0 ls26 ws0">空间点</div><div class="t m6 x60 h15 y2f ff8 fs0 fc0 sc0 ls0 ws0">A</div><div class="t m3 x61 h7 y2f ff5 fs0 fc0 sc0 ls26 ws0">与投影中心</div><div class="t m6 x62 h15 y2f ff8 fs0 fc0 sc0 ls0 ws0">O</div><div class="t m3 x63 h7 y2f ff5 fs0 fc0 sc0 ls26 ws0">连接，形成直线</div><div class="t m6 x64 h15 y2f ff8 fs0 fc0 sc0 ls2f ws0">OA</div><div class="t m3 x65 h7 y2f ff5 fs0 fc0 sc0 ls26 ws0">，并且与半球面</div><div class="t m6 x55 h15 y58 ff8 fs0 fc0 sc0 ls0 ws0">S</div><div class="t m3 x66 h7 y58 ff5 fs0 fc0 sc0 ls2b ws0">交于点</div><div class="t m6 x67 h15 y58 ff8 fs0 fc0 sc0 ls0 ws0">B</div><div class="t m3 x68 h7 y58 ff5 fs0 fc0 sc0 ls0 ws0">；</div><div class="t m5 x69 h2 y58 ff1 fs0 fc0 sc0 ls2d ws0">(2)</div><div class="t m3 x1b h7 y58 ff5 fs0 fc0 sc0 ls0 ws0">将</div><div class="t m6 x56 h15 y58 ff8 fs0 fc0 sc0 ls0 ws0">B</div><div class="t m3 x14 h7 y58 ff5 fs0 fc0 sc0 ls2b ws0">线性映射到图像平面，如图</div><div class="t m5 x6a h2 y58 ff1 fs0 fc0 sc0 ls0 ws0">1</div><div class="t m3 x6b h7 y58 ff5 fs0 fc0 sc0 ls2b ws0">所示。</div><div class="t m5 x6c h2 y58 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m7 x6d h16 y59 ff8 fs9 fc1 sc0 ls0 ws0">A</div><div class="t m7 x6e h16 y5a ff8 fs9 fc1 sc0 ls0 ws0">B</div><div class="t m7 x6f h16 y5b ff8 fs9 fc1 sc0 ls0 ws0">C</div><div class="t m7 x70 h16 y5c ff8 fs9 fc1 sc0 ls0 ws0">S</div><div class="t m7 x71 h16 y5d ff8 fs9 fc1 sc0 ls0 ws0">O</div><div class="t m7 x72 h16 y5e ff8 fs9 fc1 sc0 ls30 ws0">Z </div><div class="t m7 x73 h16 y5f ff8 fs9 fc1 sc0 ls0 ws0">Y</div><div class="t m7 x74 h16 y60 ff8 fs9 fc1 sc0 ls0 ws0">X</div><div class="t m5 x75 h2 y61 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m8 x5a ha y62 ff5 fs5 fc0 sc1 ls0 ws0">图</div><div class="t m9 x76 h17 y62 ff3 fs5 fc0 sc0 ls31 ws1e">1 </div><div class="t m8 x6f ha y62 ff5 fs5 fc0 sc1 ls32 ws0">鱼眼图像成像模型</div><div class="t m9 x77 h17 y62 ff3 fs5 fc0 sc0 ls0 ws0"> </div><div class="t m3 x59 h7 y63 ff5 fs0 fc0 sc0 ls26 ws0">假设把鱼眼相机放在坐标原点</div><div class="t m6 x78 h15 y63 ff8 fs0 fc0 sc0 ls0 ws0">O</div><div class="t m3 x79 h7 y63 ff5 fs0 fc0 sc0 ls26 ws0">，<span class="_ _7"></span>拍摄方向为</div><div class="t m6 x7a h15 y63 ff8 fs0 fc0 sc0 ls2f ws0">OZ</div><div class="t m3 x7b h7 y63 ff5 fs0 fc0 sc0 ls26 ws0">方向，</div><div class="t m3 x55 h7 y64 ff5 fs0 fc0 sc0 ls26 ws0">则像平面为</div><div class="t m6 x7c h15 y64 ff8 fs0 fc0 sc0 ls2f ws0">OXY</div><div class="t m3 xd h7 y64 ff5 fs0 fc0 sc0 ls0 ws0">。</div><div class="t m6 x7d h15 y64 ff8 fs0 fc0 sc0 ls0 ws0">S</div><div class="t m3 x7e h7 y64 ff5 fs0 fc0 sc0 ls26 ws0">是以点</div><div class="t m6 x7f h15 y64 ff8 fs0 fc0 sc0 ls0 ws0">O</div><div class="t m3 x70 h7 y64 ff5 fs0 fc0 sc0 ls26 ws0">为圆心的半圆，半径为</div><div class="t m6 x80 h15 y64 ff8 fs0 fc0 sc0 ls0 ws0">r</div><div class="t m3 x81 h7 y64 ff5 fs0 fc0 sc0 ls2b ws0">。连</div><div class="t m3 x55 h7 y65 ff5 fs0 fc0 sc0 ls0 ws0">接</div><div class="t m6 x82 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">A</div><div class="t m3 x83 h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">与</div><div class="t m6 x60 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">O</div><div class="t m3 x84 h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">交</div><div class="t m6 x69 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">S</div><div class="t m3 x1a h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">于</div><div class="t m6 xe h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">B</div><div class="t m3 x7d h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">，<span class="_ _a"></span>过</div><div class="t m6 x85 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">B</div><div class="t m3 x6f h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">作</div><div class="t m6 x86 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">Z</div><div class="t m3 x87 h7 y66 ff5 fs0 fc0 sc0 ls26 ws0">轴的平行线，<span class="_ _3"></span><span class="ls0">交</span></div><div class="t m6 x88 h15 y66 ff8 fs0 fc0 sc0 ls33 ws0">OXY</div><div class="t m3 x6b h7 y66 ff5 fs0 fc0 sc0 ls26 ws0">平面于</div><div class="t m6 x89 h15 y66 ff8 fs0 fc0 sc0 ls0 ws0">C</div><div class="t m3 x8a h7 y66 ff5 fs0 fc0 sc0 ls0 ws0">，</div><div class="t m6 x55 h15 y67 ff8 fs0 fc0 sc0 ls0 ws0">C</div><div class="t m3 x8b h7 y67 ff5 fs0 fc0 sc0 ls26 ws0">即是鱼眼镜头拍摄所成的像。当拍摄景物足够</div><div class="t m5 x8c h2 y67 ff1 fs0 fc0 sc0 ls28 ws0">180º</div><div class="t m3 x5f h7 y67 ff5 fs0 fc0 sc0 ls26 ws0">时，景</div><div class="t m3 x55 h7 y68 ff5 fs0 fc0 sc0 ls2b ws0">物成的像布满圆</div><div class="t m6 xd h15 y68 ff8 fs0 fc0 sc0 ls0 ws0">O</div><div class="t m3 x56 h7 y68 ff5 fs0 fc0 sc0 ls2b ws0">，所以，鱼眼图像是一个圆形图像。</div><div class="t m5 x8d h2 y68 ff1 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m1 x55 h4 y69 ff3 fs1 fc0 sc0 ls23 ws1d">3 </div><div class="t m2 x57 h13 y69 ff5 fs1 fc0 sc1 ls24 ws0">鱼眼图像轮廓提取</div><div class="t m1 x58 h4 y69 ff3 fs1 fc0 sc0 ls0 ws0"> </div><div class="t ma x55 h9 y6a ff3 fs0 fc0 sc0 ls2d ws1f">3.1 </div><div class="t mb x8e h7 y6a ff5 fs0 fc0 sc1 ls34 ws0">算法原理</div><div class="t ma x1a h9 y6a ff3 fs0 fc0 sc0 ls0 ws0"> </div><div class="t m3 x59 h7 y6b ff5 fs0 fc0 sc0 ls26 ws0">鱼眼图像的圆形有<span class="_ _5"></span>效区含有全部景物信息，而无效<span class="_ _5"></span>区域</div><div class="t m3 x55 h7 y6c ff5 fs0 fc0 sc0 ls26 ws0">灰度值较小，所以<span class="_ _5"></span>，采用阈值分割，将图像二值化<span class="_ _5"></span>可以提取</div><div class="t m0 x55 ha y6d ff5 fs5 fc0 sc1 ls7 ws0">基金项目：<span class="sc0 ls0">重庆科技攻关计划基金资助项目<span class="ff1 ls35">(CSTC2009AC2057)</span><span class="ff3"> </span></span></div><div class="t m0 x55 ha y6e ff5 fs5 fc0 sc1 ls7 ws0">作者简介：<span class="_ _4"></span><span class="sc0 ls0">杨<span class="ff1 ls6"> </span>丹<span class="ff1 ls5">(1962</span>－<span class="ff1">)</span>，<span class="_ _4"></span>男，<span class="_ _1"></span>教授、<span class="_ _4"></span>博士生导<span class="_ _5"></span>师，<span class="_ _4"></span>主研方向：<span class="_ _1"></span>数</span></div><div class="t m0 x55 ha y6f ff5 fs5 fc0 sc0 ls7 ws0">字图像处理，模式识别，计算机视觉，科学与工程计算；游 <span class="_ _b"> </span>磊，</div><div class="t m0 x55 ha y70 ff5 fs5 fc0 sc0 ls0 ws0">硕士研究生；张小洪，副教授；李 <span class="_ _b"> </span>博，博士<span class="ff3"> </span></div><div class="t m0 x55 ha y71 ff5 fs5 fc0 sc1 ls7 ws0">收稿日期：<span class="ff1 sc0 ls5 ws1">2009-09-28 <span class="ff3 ls36 ws0">E-mail</span></span><span class="ls0">：<span class="ff1 sc0 ls19">nation-water@163.<span class="_ _1"></span>com<span class="ff3 fs0 ls0"> </span></span></span></div></div><div class="pi" data-data='{"ctm":[1.611639,0.000000,0.000000,1.611639,0.000000,0.000000]}'></div></div> </body> </html>