B4-2a-snake.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Opencv3.0以降cv2.cvが削除され、cv.SnakeImageが参照できなくなっていたので、自作関数を作っている方からコードをお借りしました。\n",
    "\n",
    "from\n",
    "https://qiita.com/kinketsucom/items/3d48f8705cde44d88eaf"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#coding:utf-8\n",
    "import numpy as np\n",
    "import cv2\n",
    "import math\n",
    "import copy\n",
    "\n",
    "\n",
    "#画像の読み込み\n",
    "test = cv2.imread(\"images/original.png\", cv2.IMREAD_COLOR)#BGRなので気をつける\n",
    "\n",
    "gray_test = cv2.imread(\"images/original.png\",cv2.IMREAD_GRAYSCALE)\n",
    "height = test.shape[0]\n",
    "width = test.shape[1] \n",
    "#画像の書き出し\n",
    "cv2.imwrite('images/test-snake.jpg', test)\n",
    "cv2.imwrite('images/gray_test-snake.jpg',gray_test)\n",
    "\n",
    "N = 2000\n",
    "\n",
    "v = np.zeros((N,2))\n",
    "start_v = np.zeros((N,2))\n",
    "vec_g = np.zeros(2)\n",
    "for i in range(0,N):\n",
    "    if(i<N/4):\n",
    "        v[i] = [height/(N/4)*i,0]\n",
    "    elif(i<2*N/4):\n",
    "        v[i] = [height-1,width/(N/4)*(i-N/4)]\n",
    "    elif(i<3*N/4):\n",
    "        v[i] = [height-1-height/(N/4)*(i-2*N/4),width-1]\n",
    "    else:\n",
    "        v[i] = [0,width-1 - width/(N/4)*(i-3*N/4)]\n",
    "\n",
    "\n",
    "#初期輪郭点が円形の場合はこっち\n",
    "# for i in range(0,N):\n",
    "#     v[i] = [ height/2*math.sin(2*math.pi*i/N)+height/2, width/2*math.cos(2*math.pi*i/N)+width/2]\n",
    "#     start_v[i] = [ height/2*math.sin(2*math.pi*i/N)+height/2, width/2*math.cos(2*math.pi*i/N)+width/2]\n",
    "\n",
    "\n",
    "start_v = copy.deepcopy(v)\n",
    "display = copy.deepcopy(test)\n",
    "\n",
    "#パラメータ\n",
    "alpha =1\n",
    "beta = 1\n",
    "gamma = 10\n",
    "kappa =1\n",
    "\n",
    "def EpsIn(vec0,vec1,vec2):#test\n",
    "    value = 0\n",
    "    value += alpha*np.linalg.norm(vec1-vec0)**2+beta*np.linalg.norm(vec2-2*vec1+vec0)**2\n",
    "    value /= 2\n",
    "#     print(\"In:\"+str(value))\n",
    "    return value\n",
    "\n",
    "def EpsEx(vec0,pix):#gray\n",
    "    value = 0\n",
    "    x = int(vec0[0])\n",
    "    y = int(vec0[1])\n",
    "\n",
    "    if(x+1 >= height or y+1 >= width):\n",
    "        return float('inf') \n",
    "    else:\n",
    "        I = [abs(int(pix[x+1,y]) - int(pix[x,y])) ,abs(int(pix[x,y+1])-int(pix[x,y]))]\n",
    "        value = -gamma*np.linalg.norm(I)**2\n",
    "#         print(\"Ex:\"+str(value))\n",
    "        return value\n",
    "\n",
    "def EpsCon(vec0,vec_g):#test\n",
    "    value = 0\n",
    "    value += kappa*np.linalg.norm((vec0[0] - vec_g[0],vec0[1]-vec_g[1]))**2\n",
    "#     print(\"Con:\"+str(value))\n",
    "    return value\n",
    "\n",
    "def Energy(vec0,vec1,vec2,vec_g,pix):\n",
    "    value = 0\n",
    "    value = EpsIn(vec0,vec1,vec2)+EpsEx(vec0,pix)+EpsCon(vec0,vec_g)\n",
    "#     print(\"Result:\"+str(value))\n",
    "    return value\n",
    "#探索\n",
    "n = 500\n",
    "dx = [-1, -1, -1, 0, 0, 0, 1, 1, 1]\n",
    "dy = [1, 0, -1, 1, 0, -1, 1, 0, -1]\n",
    "# dx = [1,1,1,0,0,0,-1,-1,-1]\n",
    "# dy = [1,-1,0,1,-1,0,1,-1,0]\n",
    "#210\n",
    "#543\n",
    "#876\n",
    "\n",
    "flag = 4\n",
    "for loop in range(0,n):\n",
    "    for i in range(0,N):\n",
    "        flag = 4\n",
    "        eps_min = float('inf') \n",
    "        vec_g = [0,0]\n",
    "\n",
    "        #重心中心にするならこれ\n",
    "#         for j in range(0,N):\n",
    "#             vec_g += [v[j,0],v[j,1]]\n",
    "\n",
    "        for j in range(0,9):            \n",
    "            move  = [v[i,0]+dx[j], v[i,1]+dy[j]]\n",
    "            if(move[0] < 0 or move[1] < 0 or move[0] >= height  or move[1] >= width):\n",
    "                continue #はみ出し処理\n",
    "\n",
    "            #重心中心にするならこれ\n",
    "            #vec_g += [dx[j],dy[j]]\n",
    "            #vec_g =[vec_g[0]/N, vec_g[1]/N]\n",
    "            #画像中心を基準に\n",
    "            vec_g = [int(height/2),int(width/2)]\n",
    "\n",
    "            energy = Energy(move,v[(i+1)%N],v[(i+2)%N],vec_g,gray_test)\n",
    "            if(eps_min>energy):\n",
    "                eps_min = energy\n",
    "                flag = j\n",
    "        v[i] += [dx[flag],dy[flag]]\n",
    "\n",
    "        #逐次書き出し\n",
    "    if(loop%10==0):\n",
    "        cv2.imwrite('images/Snake-result'+str(loop)+'.jpg', display)\n",
    "        display = copy.deepcopy(test)\n",
    "        for i in range(0,N):\n",
    "            cv2.line(display, (int(v[i,1]),int(v[i,0])), (int(v[(i+1)%N,1]),int(v[(i+1)%N,0])), (0, 255, 0), 2)\n",
    "\n",
    "\n",
    "\n",
    "for i in range(0,N):\n",
    "    cv2.line(display, (int(v[i,1]),int(v[i,0])), (int(v[(i+1)%N,1]),int(v[(i+1)%N,0])), (0, 255, 0), 2)\n",
    "\n",
    "for i in range(0,N):\n",
    "    cv2.line(display, (int(start_v[i,1]),int(start_v[i,0])), (int(start_v[(i+1)%N,1]),int(start_v[(i+1)%N,0])), (255, 0, 0), 2)\n",
    "    \n",
    "cv2.imwrite('images/result-snake.jpg', display)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from PIL import Image\n",
    "\n",
    "images = []\n",
    "\n",
    "for i in range(0,490,10):\n",
    "    file_name = 'images/Snake-result'+str(i)+'.jpg'\n",
    "    im = Image.open(file_name)\n",
    "    images.append(im)\n",
    "\n",
    "images[0].save('snake.gif', save_all=True, append_images=images[1:], loop=0, duration=30)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![SegmentLocal](images/snake.gif \"segment\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}