openc-Python扫描照片中的文档

文档扫描预览

将图片中的档变成易于查看的扫描件

当前图片：

处理后的图片：

具体实现

利用Canny算子对图片进行处理

利用Canny算子对图片进行边缘检测，得到一个二值化相对较好的图片，滤除背景颜色。

#滤波
blur_picture = cv2.GaussianBlur(picture_for_scan, (5, 5), 0)
cv_show('blur_picture', blur_picture)

#canny算子，边缘检测（双阈值处理）
edged_picture = cv2.Canny(blur_picture, 50, 200)
ret , edged_picture = cv2.threshold(edged_picture, 200, 255, cv2.THRESH_BINARY) 
cv_show('edged_picture', edged_picture)

边缘处理后的图片：

对处理后图片进行轮廓检测

contours , hes = cv2.findContours(edged_picture.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
draw_contours_picture = cv2.drawContours(color_picture, contours, -1, (0, 255, 0), 2)
cv_show('draw_contours_picture', draw_contours_picture)

轮廓检测效果：

筛选轮廓并拉直为四边形

因为检测出的轮廓可能是多个，我们选择面积最大的那个轮廓框架。

if len(contours) > 0:
    #按面积大小排序
    contours = sorted(contours, key = cv2.contourArea , reverse = True)

选出最大轮廓后，因为轮廓可能是个多边形，所以需要把它拉直为四边形。

#将轮廓拉直为四边形   
c = contours[0]
peri = cv2.arcLength(c, True)
for k in np.arange(0,0.1,0.01):
    approx = cv2.approxPolyDP(c, k*peri, True)
    if len(approx) == 4:
        docCnt = approx
        break

对图像进行透视变换

对轮廓的四角坐标进行排序

将四个角坐标进行重排，重排顺序为：左上， 右上， 左下， 右下。

#将四个角坐标排序：左上， 右上， 左下， 右下
def order_points(pts):
	#按x坐标重排
    sort_id = np.argsort(pts[:,0])
    pts = pts[sort_id, :]
    #按y坐标重排
    mid_sort_id = np.argsort(pts[0:2, 1])
    pts[0:2, :] = pts[mid_sort_id, :]
    mid_sort_id = np.argsort(pts[2:4, 1])
    mid_pts = pts[2:4, :]
    pts[2:4 :] = mid_pts[mid_sort_id, :]
    return pts

得到变换后坐标,并变换

选取长变和最大的高作为变换后图片的长和宽。（ps：变换后坐标顺序为：左上， 右上， 左下， 右下）

rect = order_points(pts)
(tl, tr, bl, br) = rect

#计算轮廓参数
widthA = distance(br, bl)
widthB = distance(tr, tl)
maxWidth = max(int(widthA), int(widthB))

heigthA = distance(tr, br)
heigthB = distance(tl, bl)
maxHeigth = max(int(heigthA), int(heigthB))

#变换后坐标
dst = np.array([
    [0, 0],
    [0, maxWidth-1],
    [maxHeigth-1, 0],
    [maxHeigth-1, maxWidth-1]
],dtype = np.float32)
#数据类型转换
rect = rect.astype(np.float32)

坐标变换：

#计算变换矩阵
M = cv2.getPerspectiveTransform(rect, dst)
#图像坐标变换
warped = cv2.warpPerspective(img, M, (maxHeigth, maxWidth))

附录

#扫描图片

import cv2
import numpy as np

def cv_show(name,img):
    cv2.imshow(name, img)
    cv2.imwrite('./scan_picture/' + name + '.png', img)
    cv2.waitKey(0)
    cv2.destroyAllWindows()

def order_points(pts):
	#按x坐标重排
    sort_id = np.argsort(pts[:,0])
    pts = pts[sort_id, :]
    #按y坐标重排
    mid_sort_id = np.argsort(pts[0:2, 1])
    pts[0:2, :] = pts[mid_sort_id, :]
    mid_sort_id = np.argsort(pts[2:4, 1])
    mid_pts = pts[2:4, :]
    pts[2:4 :] = mid_pts[mid_sort_id, :]
    return pts

#计算两点间曼哈顿距离
def distance(x , y):
    dis = np.sqrt(((x[0]-y[0])**2)+((x[1]-y[1])**2))
    return dis

#图像坐标变换
def four_point_transform(img , pts):
    rect = order_points(pts)
    (tl, tr, bl, br) = rect

    #计算轮廓参数
    widthA = distance(br, bl)
    widthB = distance(tr, tl)
    maxWidth = max(int(widthA), int(widthB))
    
    heigthA = distance(tr, br)
    heigthB = distance(tl, bl)
    maxHeigth = max(int(heigthA), int(heigthB))
    
    #变换后坐标
    dst = np.array([
		[0, 0],
	    [0, maxWidth-1],
	    [maxHeigth-1, 0],
	    [maxHeigth-1, maxWidth-1]
	],dtype = np.float32)
    #数据类型转换
    rect = rect.astype(np.float32)
    
    #计算变换矩阵
    M = cv2.getPerspectiveTransform(rect, dst)
    warped = cv2.warpPerspective(img, M, (maxWidth, maxHeigth))
    return warped
    
picture_for_scan = cv2.imread('./scan_picture/origina_picture.png' , 0)
origina_picture = cv2.imread('./scan_picture/origina_picture.png' , 1)
color_picture = origina_picture.copy()
cv_show('origina_picture', origina_picture)

#滤波
blur_picture = cv2.GaussianBlur(picture_for_scan, (5, 5), 0)
cv_show('blur_picture', blur_picture)

#canny算子，边缘检测（双阈值处理）
edged_picture = cv2.Canny(blur_picture, 50, 200)
ret , edged_picture = cv2.threshold(edged_picture, 200, 255, cv2.THRESH_BINARY) 
cv_show('edged_picture', edged_picture)

contours , hes = cv2.findContours(edged_picture.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
draw_contours_picture = cv2.drawContours(color_picture, contours, -1, (0, 255, 0), 2)
cv_show('draw_contours_picture', draw_contours_picture)

if len(contours) > 0:
    #按面积大小排序
    contours = sorted(contours, key = cv2.contourArea , reverse = True)
  
#将轮廓拉直为四边形   
c = contours[0]
peri = cv2.arcLength(c, True)
for k in np.arange(0,0.1,0.01):
    approx = cv2.approxPolyDP(c, k*peri, True)
    if len(approx) == 4:
        docCnt = approx
        break
       
scaned_picture = four_point_transform(picture_for_scan.copy() , docCnt.reshape(4, 2))
#二值化处理，使得背景为白色
#ret , scaned_picture = cv2.threshold(scaned_picture, 0, 255, cv2.THRESH_OTSU)
cv_show('scaned_picture', scaned_picture)