区域生长法的编程实现——程序员数字图像处理第一步

数字图像处理的书数不胜数。相关的方法，从直方图、卷积到小波、机器学习方面的方法也举不胜举。 OpenCV 库给我们提供了一整套图像存储、显示方法外，也集成了很多很多的图像、视频处理算法以及机器学习算法，以函数的形式提 供给我们使用。 然而在初学阶段，我们往往过分依赖于 OpenCV 中的那些库函数。 虽说 OpenCV 集成了很多高效算法，然而，毕竟不是无所不包。 OpenCV 中基础的算法有时会缺失，有时用起来不是很得心应手。 区域生长算法就是一例。 区域生长法，思路很简单：将一个图像中的一点作为种子，此种子会沿 8 邻域/4 邻域进行扩展，扩展具有一定限制（即差值不能过大， 当然你可以引入梯度这个概念，甚至二维偏导数），求出种子生长的最终结果。 把示例程序的几张结果发过来，何谓区域生长法，大家一看便知。 区域生长法对于较为均衡的背景中提取目标物而言，效果较为理想。与阈值法、边缘检测法孰优孰劣，俺笔力及精力有限，不进行比较 了。单从工程角度，区域生长法是统计闭合图形个数的基础。 这里自己写了一下这个程序的演示代码，只负责将背景同生长不上去的区域分开： cpp view plaincopy 1. #define INIT 0 2. #define SEED -1 3. #define CONT -2 4. #define INVAL -3 5. 6. int color_templ5*6*3 7. = 8. 0,0,0, 40,40,40, 80,80,80, 120,120,120,160,160,160, 200,200,200, 9. 76,0,13, 113,52,0, 127,91,0, 188,114,0, 203,140,68, 236,211,190, 10. 0,0,121, 93,0,158, 91,20,237, 255,61,255, 170,110,240, 200,161,245, 11. 38,88,0, 77,118,46, 66,130,98, 112,182,46, 118,197,124, 174,221,178, 12. 19,57,96, 10,98,163, 0,160,171, 0,242,255, 59,245,255, 139,247,255 13. ; 14. 15. class g_point 16. 17. public: 18. g_point(); 19. g_point(int x_i, int y_i); 20. g_point(int x_i, int y_i, int lbl_i); 21. public: 22. int x; 23. int y; 24. int lbl; /标签 25. ; 26. 27. g_point:g_point(int x_i, int y_i) 28. 29. x = x_i; 30. y = y_i; 31. lbl = SEED; 32. 33. 34. g_point:g_point() 35. 36. x = 0; 37. y = 0; 38. lbl = INIT; 39. 40. 41. g_point:g_point(int x_i, int y_i, int lbl_i) 42. 43. x = x_i; 44. y = y_i; 45. lbl = lbl_i; 46. 47. 48. void regionGrowth(IplImage* img, int* p_mat, int x_g, int y_g, int threshold_g) 49. 50. /常用变量： 51. int x; 52. int y; 53. int gradient = 0; 54. 55. 56. /设置前次边缘点 57. std:list cont_pts_pre; 58. std:list * pcont_pts_pre = 59. 60. 61. /设置边缘点 62. std:list cont_pts; 63. std:list * pcont_pts = 64. 65. 66. /种子点+边缘点 = 下次的种子点 67. /上次的边缘点 = 前次边缘点（用于下次减少点数) 68. cont_pts_pre.push_back(class g_point(x_g, y_g, CONT); 69. p_maty_g * img-width + x_g = SEED; 70. 71. std:list:iterator iter_cont; 72. std:list:iterator iter_prt; 73. std:list:iterator iter_swap; 74. 75. 76. 77. 78. while( !cont_pts_pre.empty() ) 79. 80. 81. 82. 83. /一轮生长 84. iter_cont = cont_pts_pre.begin(); 85. while(iter_cont != cont_pts_pre.end() 86. 87. x = (*iter_cont).x; 88. y = (*iter_cont).y; 89. / if( !(x-1width + (x-1) = INIT) 92. / 93. / gradient = (char*)(img-imageData + y*img-widthStep)x - 94. / (char*)(img-imageData + (y-1)*img-widthStep)x-1; 95. / if(abs(gradient) width + (x-1) = SEED; 99. / 100. / else /不满足阈值要求 101. / 102. / p_mat(y-1)*img-width + (x-1) = INVAL; 103. / 104. / 105. / 106. if( !(x-1width + (x-1) = INIT) 109. 110. gradient = (char*)(img-imageData + y*img-widthStep)x - 111. (char*)(img-imageData + (y)*img-widthStep)x-1; 112. if(abs(gradient) width + (x-1) = SEED; 116. 117. else /不满足阈值要求 118. 119. p_mat(y)*img-width + (x-1) = INVAL; 120. 121. 122. 123. 124. / if( !(x-1= img-height) ) /#3 125. / 126. / if(p_mat(y+1)*img-width + (x-1) = INIT) 127. / 128. / gradient = (char*)(img-imageData + y*img-widthStep)x - 129. / (char*)(img-imageData + (y+1)*img-widthStep)x-1; 130. / if(abs(gradient) width + (x-1) = SEED; 134. / 135. / else /不满足阈值要求 136. / 137. / p_mat(y+1)*img-width + (x-1) = INVAL; 138. / 139. / 140. / 141. / 142. if( !(y+1 = img-height) ) /#4 143. 144. if(p_mat(y+1)*img-width + (x) = INIT) 145. 146. gradient = (char*)(img-imageData + y*img-widthStep)x - 147. (char*)(img-imageData + (y+1)*img-widthStep)x; 148. if(abs(gradient) width + (x) = SEED; 152. 153. else /不满足阈值要求 154. 155. p_mat(y+1)*img-width + (x) = INVAL; 156. 157. 158. 159. / if( !(x+1=img-width | y+1=img-height) ) /#5 160. / 161. / if(p_mat(y+1)*img-width + (x+1) = INIT) 162. / 163. / gradient = (char*)(img-imageData + y*img-widthStep)x - 164. / (char*)(img-imageData + (y+1)*img-widthStep)x+1; 165. / if(abs(gradient) width + (x+1) = SEED; 169. / 170. / else /不满足阈值要求 171. / 172. / p_mat(y+1)*img-width + (x+1) = INVAL; 173. / 174. / 175. / 176. / 177. if( !(x+1=img-width) ) /#6 178. 179. if(p_mat(y)*img-width + (x+1) = INIT) 180. 181. gradient = (char*)(img-imageData + y*img-widthStep)x - 182. (char*)(img-imageData + (y)*img-widthStep)x+1; 183. if(abs(gradient) width + (x+1) = SEED; 187. 188. else /不满足阈值要求 189. 190. p_mat(y)*img-width + (x+1) = INVAL; 191. 192. 193. 194. 195. / if( !(x+1=img-width | y-1width + (x+1) = INIT) 198. / 199. / gradient = (char*)(img-imageData + y*img-widthStep)x - 200. / (char*)(img-imageData + (y-1)*img-widthStep)x+1; 201. / if(abs(gradient) width + (x+1) = SEED; 205. / 206. / else /不满足阈值要求 207. / 208. / p_mat(y-1)*img-width + (x+1) = INVAL; 209. / 210. / 211. / 212. / 213. if( !(y-1width + (x) = INIT) 216. 217. gradient = (char*)(img-imageData + y*img-widthStep)x - 218. (char*)(img-imageData + (y-1)*img-widthStep)x; 219. if(abs(gradient) width + (x) = SEED; 223. 224. else /不满足阈值要求 225. 226. p_mat(y-1)*img-width + (x) = INVAL; 227. 228. 229. 230. 231. iter_cont +; 232. 233. 234. /将 cont_pts 中的点赋给 cont_pts_pre 235. cont_pts_pre.clear(); 236. iter_swap = cont_pts.begin(); 237. while(iter_swap != cont_pts.end() ) 238. 239. cont_pts_pre.push_back(*iter_swap); 240. iter_swap +; 241. 242. cont_pts.clear(); 243. 244. 245. 246. /结束 while 247. 248. 249. 250. 251. cont_pts_pre.clear(); 252. cont_pts.clear(); 253. 254. / delete pseed_pts; 255. / delete pcont_pts_pre; 256. / delete pcont_pts; 257. 258. 259. 260. void display_mat(int * p_mat, int width, int height) 261. 262. IplImage* disp; 263. disp = cvCreateImage( cvSize(width, height), 8, 3 ); 264. cvZero(disp); 265. 266. for(int h = 0; h imageData + h*disp-widthStep)w*3 + 0 = color_templ0*18 + 0*3 + 0; 273. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 1 = color_templ0*18 + 0*3 + 1; 274. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 2 = color_templ0*18 + 0*3 + 2; 275. 276. else if(p_math * width + w = INIT) 277. 278. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 0 = color_templ0*18 + 2*3 + 0; 279. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 1 = color_templ0*18 + 2*3 + 1; 280. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 2 = color_templ0*18 + 2*3 + 2; 281. 282. else if(p_math * width + w = INVAL) 283. 284. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 0 = color_templ0*18 + 4*3 + 0; 285. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 1 = color_templ0*18 + 4*3 + 1; 286. (UCHAR *)(disp-imageData + h*disp-widthStep)w*3 + 2 = color_templ0*18 + 4*3 + 2; 287. 288. 289. 290. cvNamedWindow(“display“); 291. cvShowImage(“display“, disp); 292. cvWaitKey(0); 293. 294. cvDestroyWindow(“display“); 295. cvReleaseImage( 296. 297. 298. 299. int _tmain(int argc, _TCHAR* argv) 300. 301. IplImage* img; 302. img = cvLoadImage(“1.bmp“, CV_LOAD_IMAGE_ANYCOLOR); 303. 304. 305. /建立需要使用的单通道 8 位深图像 306. IplImage* img_gray; 307. img_gray = cvCreateImage(cvGetSize(img), img-depth, 1); 308. cvSplit(img, NULL, img_gray, NULL, NULL); 309. 310. cvNamedWindow(“original image“, -1); 311. cvShowImage(“original image“, img_gray); 312. cvWaitKey(0); 313. 314. /建立所需的与图像同样宽高的整型数矩阵并赋初值 315. int * pres_mat; 316. pres_mat = new intimg_gray-width * img_gray-height; 317. for(int i = 0; i width * img_gray-height; i+) 318. 319. pres_mati = INIT; 320. 321. 322. /开始进行区域生长 323. regionGrowth(img_gray, pres_mat, 1, 1, 50); 324. 325. display_mat(pres_mat, img_gray-width, img_gray-height); 326. 327. return 0; 328. regionGrowth 函数详解 参数： img：需要进行区域生长的图像，8 位单通道的 IplImage p_map：整型数矩阵指针，宽高与图像相同，用于储存图像生长结果 矩阵已经被全部赋值为 INIT(#define INIT 0) x_g, y_g 区域生长初始种子点