传统opencv分割多张票据

针对于票据或者是多个零配件等物体通过修改对应参数可以做到大部分近似矩形物体的分割

由于无法上传.ipynb文件，各位可以复制进jupyter内查看并debug代码

### 0.导入包 定义显示函数

1. {
   
2. "cells": [
   
3.   {
   
4.    "cell_type": "markdown",
   
5.    "metadata": {},
   
6.    "source": [
   
7.     "### 0.导入包 定义显示函数"
   
8.    ]
   
9.   },
   
10.   {
    
11.    "cell_type": "code",
    
12.    "execution_count": null,
    
13.    "metadata": {},
    
14.    "outputs": [],
    
15.    "source": [
    
16.     "import cv2 as cv\n",
    
17.     "import numpy as np\n",
    
18.     "\n",
    
19.     "def show_img(img, win_name):\n",
    
20.     "\n",
    
21.     "    # cv.namedWindow(win_name, cv.WINDOW_NORMAL | cv.WINDOW_KEEPRATIO)\n",
    
22.     "\n",
    
23.     "    cv.imshow(win_name, img)\n",
    
24.     "    \n",
    
25.     "    cv.waitKey(0)\n",
    
26.     "    cv.destroyAllWindows()"
    
27.    ]
    
28.   },
    
29.   {
    
30.    "cell_type": "markdown",
    
31.    "metadata": {},
    
32.    "source": [
    
33.     "### 1.读取图片 转为灰度图 利用高斯模糊消除噪声"
    
34.    ]
    
35.   },
    
36.   {
    
37.    "cell_type": "code",
    
38.    "execution_count": null,
    
39.    "metadata": {},
    
40.    "outputs": [],
    
41.    "source": [
    
42.     "img_path = './fapiao_test//2tickets_test3.jpg'\n",
    
43.     "img = cv.imread(img_path)\n",
    
44.     "img = cv.resize(img, (600, 900))\n",
    
45.     "show_img(img, 'img')\n",
    
46.     "\n",
    
47.     "gray = cv.cvtColor(img, cv.COLOR_RGB2GRAY)\n",
    
48.     "\n",
    
49.     "gray = cv.GaussianBlur(gray, (7, 7), 0)\n",
    
50.     "show_img(gray, 'gray')"
    
51.    ]
    
52.   },
    
53.   {
    
54.    "cell_type": "markdown",
    
55.    "metadata": {},
    
56.    "source": [
    
57.     "### 2.使用边缘检测"
    
58.    ]
    
59.   },
    
60.   {
    
61.    "cell_type": "code",
    
62.    "execution_count": null,
    
63.    "metadata": {},
    
64.    "outputs": [],
    
65.    "source": [
    
66.     "edged = cv.Canny(gray, 100, 200)\n",
    
67.     "show_img(edged, 'edged')"
    
68.    ]
    
69.   },
    
70.   {
    
71.    "cell_type": "markdown",
    
72.    "metadata": {},
    
73.    "source": [
    
74.     "### 3.形态学操作 进行腐蚀膨胀进一步消除噪声"
    
75.    ]
    
76.   },
    
77.   {
    
78.    "cell_type": "code",
    
79.    "execution_count": null,
    
80.    "metadata": {},
    
81.    "outputs": [],
    
82.    "source": [
    
83.     "kernel = np.ones((7, 7), np.uint8)\n",
    
84.     "morphed = cv.dilate(edged, kernel, iterations=3)\n",
    
85.     "morphed = cv.erode(morphed, kernel, iterations=3)\n",
    
86.     "show_img(morphed, 'morphed')"
    
87.    ]
    
88.   },
    
89.   {
    
90.    "cell_type": "markdown",
    
91.    "metadata": {},
    
92.    "source": [
    
93.     "### 4.寻找轮廓 并排序 找出其中最大的4个轮廓"
    
94.    ]
    
95.   },
    
96.   {
    
97.    "cell_type": "code",
    
98.    "execution_count": null,
    
99.    "metadata": {},
    
100.    "outputs": [],
     
101.    "source": [
     
102.     "morphed_copy = morphed.copy()\n",
     
103.     "cnts, _ = cv.findContours(morphed_copy, cv.RETR_TREE, cv.CHAIN_APPROX_NONE)\n",
     
104.     "if len(cnts) is not 0:\n",
     
105.     "    cnts = sorted(cnts, key=cv.contourArea, reverse=True)[:4]\n",
     
106.     "    print(len(cnts))\n",
     
107.     "else:\n",
     
108.     "    print("Did not find contours\\n")\n",
     
109.     "\n",
     
110.     "# for cnt in cnts:\n",
     
111.     "#     x, y, w, h = cv.boundingRect(cnt)\n",
     
112.     "#     print(x, y, w, h)\n",
     
113.     "#     print('box_area: ', w * h)\n"
     
114.    ]
     
115.   },
     
116.   {
     
117.    "cell_type": "markdown",
     
118.    "metadata": {},
     
119.    "source": [
     
120.     "### 5.将轮廓处理成矩形 在原图上裁剪"
     
121.    ]
     
122.   },
     
123.   {
     
124.    "cell_type": "code",
     
125.    "execution_count": null,
     
126.    "metadata": {},
     
127.    "outputs": [],
     
128.    "source": [
     
129.     "for cnt in cnts:\n",
     
130.     "    # 用周长的0.1倍作为阈值，对轮廓做近似处理，使其变成一个矩形\n",
     
131.     "    epsilon = 0.1 * cv.arcLength(cnt, True)\n",
     
132.     "    approx = cv.approxPolyDP(cnt, epsilon, True)\n",
     
133.     "\n",
     
134.     "    ticket_copy = img.copy()\n",
     
135.     "    cv.drawContours(ticket_copy, [approx], -1, (255, 0, 0), 2)\n",
     
136.     "    show_img(ticket_copy, 'ticket_copy')\n",
     
137.     "\n",
     
138.     "    # 获取透视变换的原坐标\n",
     
139.     "    if approx.shape[0] is not 4:\n",
     
140.     "        print("Found a non-rect\\n")\n",
     
141.     "        continue\n",
     
142.     "    src_coor = np.reshape(approx, (4, 2))\n",
     
143.     "    src_coor = np.float32(src_coor)\n",
     
144.     "\n",
     
145.     "    # 右上,左上,左下,右下 坐标\n",
     
146.     "    (tr, tl, bl, br) = src_coor\n",
     
147.     "    print(tl[0], tl[1], br[0], br[1])\n",
     
148.     "\n",
     
149.     "    x1,x2 = min(tl[0], br[0]), max(tl[0], br[0])\n",
     
150.     "    y1,y2 = min(tl[1], br[1]), max(tl[1], br[1])\n",
     
151.     "\n",
     
152.     "    img_result = img[int(y1):int(y2), int(x1):int(x2)]\n",
     
153.     "    show_img(img_result, 'img_result')\n",
     
154.     "\n",
     
155.     "    # # 计算宽\n",
     
156.     "    # w1 = np.sqrt((tr[0] - tl[0]) ** 2 + (tr[1] - tl[1]) ** 2)\n",
     
157.     "    # w2 = np.sqrt((br[0] - bl[0]) ** 2 + (br[1] - bl[1]) ** 2)\n",
     
158.     "    # # 求出比较大的w\n",
     
159.     "    # max_w = max(int(w1), int(w2))\n",
     
160.     "    # # 计算高\n",
     
161.     "    # h1 = np.sqrt((bl[0] - tl[0]) ** 2 + (bl[1] - tl[1]) ** 2)\n",
     
162.     "    # h2 = np.sqrt((br[0] - tr[0]) ** 2 + (br[1] - tr[1]) ** 2)\n",
     
163.     "    # # 求出比较大的h\n",
     
164.     "    # max_h = max(int(h1), int(h2))\n",
     
165.     "\n",
     
166.     "    # # 透视变换的目标坐标\n",
     
167.     "    # dst_coor = np.array([[max_w - 1, 0], [0, 0], [0, max_h - 1], [max_w - 1, max_h - 1]], dtype=np.float32)\n",
     
168.     "\n",
     
169.     "    # # 求转换矩阵\n",
     
170.     "    # trans_mat = cv.getPerspectiveTransform(src_coor, dst_coor)\n",
     
171.     "    # # 进行转换，将图中对应坐标的图片截取出来，并转换到dst_coor大小\n",
     
172.     "    # warped = cv.warpPerspective(img, trans_mat, (max_w, max_h))\n",
     
173.     "\n",
     
174.     "    # warped = cv.rotate(warped, cv.ROTATE_90_COUNTERCLOCKWISE)\n",
     
175.     "\n",
     
176.     "    # show_img(warped, 'result')"
     
177.    ]
     
178.   }
     
179. ],
     
180. "metadata": {
     
181.   "kernelspec": {
     
182.    "display_name": "yolo-pose",
     
183.    "language": "python",
     
184.    "name": "python3"
     
185.   },
     
186.   "language_info": {
     
187.    "codemirror_mode": {
     
188.     "name": "ipython",
     
189.     "version": 3
     
190.    },
     
191.    "file_extension": ".py",
     
192.    "mimetype": "text/x-python",
     
193.    "name": "python",
     
194.    "nbconvert_exporter": "python",
     
195.    "pygments_lexer": "ipython3",
     
196.    "version": "3.7.12"
     
197.   },
     
198.   "orig_nbformat": 4
     
199. },
     
200. "nbformat": 4,
     
201. "nbformat_minor": 2
     
202. }
     

复制代码