【OpenCV】 ⚠️实战⚠️ 女子深夜久久不能入眠，300行写出全能扫描王! ☢️建议手收藏☢️

概述图像透视获取透视矩阵透视变换 预处理其他函数主函数输出结果最终转换结果

概述

今天带大家使用我们之前学会的知识来实现一个简易版的全能扫描王. 代码分为 3 个部分: 主函程序, 预处理, 其他程序.

图像透视

透视变换 (Perspective Transformation) 是将成像投影到一个新的视平面 (Viewing Plane). 通过改变图像或视屏的透视图, 我们可以更好的了解所需信息. 如图:

获取透视矩阵

格式:

M = cv2.getPerspectiveTransform(src, dst, solveMethod=None)

参数:

src: 源图像中四边形顶点的坐标

dst: 目标图像中相应四边形顶点的坐标

返回值:

M: 透视变换矩阵

矩阵例子:

[[ 8.75046884e+00 -9.23660638e+00 0.00000000e+00][ 4.43412734e-01 -6.70045886e+00 2.19700000e+03][ 2.51096074e-03 -3.35871928e-03 1.00000000e+00]]

透视变换

格式:

cv2.warpPerspective(src, M, dsize, dst=None, flags=None, borderMode=None, borderValue=None)1

参数:

src: 输入图片M: 透视变换矩阵dsize: 输出图片的大小

返回值:

透视变换后的图片

例子:

预处理

通过图片伸缩, 灰度转换, 高斯滤波, 边缘检测等方法, 实现图片预处理

代码:

import cv2from matplotlib import pyplot as pltfrom my_functions import resizefrom my_functions import four_point_transformdef read_image(image_path, visualize=False):"""读取图片:param image_path: 图片路径:param visualize: 可视化, 默认为False:return: 返回原始图片, 裁剪后的图片, 边缘, 图片伸缩比例"""# 读取图片image = cv2.imread(image_path)# 计算伸缩比例ratio = image.shape[0] / 500.0# 深拷贝image_copy = image.copy()# 缩放大小image_resize = resize(image_copy, height=500)# 转换成灰度图image_gray = cv2.cvtColor(image_resize, cv2.COLOR_BGR2GRAY)# 高斯滤波image_gaussian = cv2.GaussianBlur(image_gray, (5, 5), 0)# Canny边缘检测edge = cv2.Canny(image_gaussian, 75, 200)if visualize:"""图片展示"""# 绘制子图f, ax = plt.subplots(2, 2, figsize=(8, 10))ax[0, 0].imshow(cv2.cvtColor(image_resize, cv2.COLOR_BGR2RGB))ax[0, 1].imshow(image_gray, "gray")ax[1, 0].imshow(image_gaussian, "gray")ax[1, 1].imshow(edge, "gray")# 去除x, y坐标ax[0, 0].set_xticks([])ax[0, 0].set_yticks([])ax[0, 1].set_xticks([])ax[0, 1].set_yticks([])ax[1, 0].set_xticks([])ax[1, 0].set_yticks([])ax[1, 1].set_xticks([])ax[1, 1].set_yticks([])# 标题ax[0, 0].set_title("original")ax[0, 1].set_title("image gray")ax[1, 0].set_title("image gaussian blur")ax[1, 1].set_title("image edge")plt.show()# 返回return image, image_resize, edge, ratiodef image_calculate_contours(image_resize, edge, visualize=False):"""计算轮廓:param image_resize: 裁剪后的图片:param edge: 边缘:param visualize: 可视化, 默认为False:return: 外接长方形数组"""# 轮廓检测contours, hierarchy = cv2.findContours(edge.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)# 轮廓排序contours = sorted(contours, key=cv2.contourArea, reverse=True)# 取最大的5个contours_5 = contours[:5]# 长方形数字approx_array = []# 遍历轮廓for c in contours_5:# 计算轮廓周长length = cv2.arcLength(c, True)# 计算大约轮廓approx = cv2.approxPolyDP(c, 0.02 * length, True)# 如果是四边形, 取出if len(approx) == 4:approx_array.append(approx)if visualize:"""图片展示"""# 绘制轮廓1draw_img_1 = cv2.drawContours(image_resize.copy(), contours, 0, (0, 255, 255), 2)# 绘制轮廓2draw_img_2 = cv2.drawContours(image_resize.copy(), contours_5, 0, (0, 255, 255), 2)# 绘制轮廓3draw_img_3 = cv2.drawContours(image_resize.copy(), approx_array, 0, (0, 255, 255), 2)# 绘制子图f, ax = plt.subplots(2, 2, figsize=(8, 10))ax[0, 0].imshow(cv2.cvtColor(image_resize, cv2.COLOR_BGR2RGB))ax[0, 1].imshow(cv2.cvtColor(draw_img_1, cv2.COLOR_BGR2RGB))ax[1, 0].imshow(cv2.cvtColor(draw_img_2, cv2.COLOR_BGR2RGB))ax[1, 1].imshow(cv2.cvtColor(draw_img_3, cv2.COLOR_BGR2RGB))# 去除x, y坐标ax[0, 0].set_xticks([])ax[0, 0].set_yticks([])ax[0, 1].set_xticks([])ax[0, 1].set_yticks([])ax[1, 0].set_xticks([])ax[1, 0].set_yticks([])ax[1, 1].set_xticks([])ax[1, 1].set_yticks([])# 标题ax[0, 0].set_title("original")ax[0, 1].set_title("contours")ax[1, 0].set_title("contours biggest 5")ax[1, 1].set_title("contours approx")plt.show()# 返回return approx_arraydef image_transform(image, approx_array, ratio, visualize=False):"""图片转换:param image: 原始图像:param approx_array: 外接长方形坐标数组:param ratio: 图片拉伸比例:param visualize: 可视化, 默认为False:return: 返回最终结果"""# 透视变换warped = four_point_transform(image, approx_array[0].reshape(4, 2) * ratio)# 转换为灰度图warped_gray = cv2.cvtColor(warped, cv2.COLOR_BGR2GRAY)# 二值化ret, thresh = cv2.threshold(warped_gray, 100, 255, cv2.THRESH_BINARY)if visualize:"""图片展示"""# 绘制子图f, ax = plt.subplots(2, 2, figsize=(8, 10))ax[0, 0].imshow(cv2.cvtColor(image, cv2.COLOR_BGR2RGB))ax[0, 1].imshow(cv2.cvtColor(warped, cv2.COLOR_BGR2RGB))ax[1, 0].imshow(warped_gray, "gray")ax[1, 1].imshow(thresh, "gray")# 去除x, y坐标ax[0, 0].set_xticks([])ax[0, 0].set_yticks([])ax[0, 1].set_xticks([])ax[0, 1].set_yticks([])ax[1, 0].set_xticks([])ax[1, 0].set_yticks([])ax[1, 1].set_xticks([])ax[1, 1].set_yticks([])# 标题ax[0, 0].set_title("original")ax[0, 1].set_title("warped")ax[1, 0].set_title("warped gray")ax[1, 1].set_title("thresh")plt.show()# 返回return warped_gray

其他函数

import numpy as npimport cv2def order_points(points):"""坐标点排序:param points: 轮廓坐标:return: 返回排序完的坐标"""# 一共4个坐标点， 左上, 右上, 右下, 左下rect = np.zeros((4, 2), dtype=np.float32)# 计算左上, 右下s = points.sum(axis=1)rect[0] = points[np.argmin(s)] # 和最小的是左上rect[2] = points[np.argmax(s)] # 和最大的是右下# 计算右上, 左下diff = np.diff(points, axis=1)rect[1] = points[np.argmin(diff)] # 差最小的是右上rect[3] = points[np.argmax(diff)] # 差最小的是左下# 返回return rectdef four_point_transform(image, pts):"""透视变换 (拉伸为长方形):param image: 原始图像:param pts: 坐标点:return: 透视变换后的图"""# 获取输入坐标点rect = order_points(pts)(top_left, top_right, bottom_left, bottom_right) = rect# 计算最大的w (勾股定理w = (Δx^2 + Δy^2)^1/2)widthA = np.sqrt(((bottom_right[0] - bottom_left[0]) ** 2) + ((bottom_right[1] - bottom_left[1]) ** 2))widthB = np.sqrt(((top_right[0] - top_left[0]) ** 2) + ((top_right[1] - top_left[1]) ** 2))maxWidth = max(int(widthA), int(widthB))# 计算最大的y (勾股定理y = (Δx^2 + Δy^2)^1/2)heightA = np.sqrt(((top_right[0] - bottom_right[0]) ** 2) + ((top_right[1] - bottom_right[1]) ** 2))heightB = np.sqrt(((top_left[0] - bottom_left[0]) ** 2) + ((top_right[1] - top_left[1]) ** 2))maxHeight = max(int(heightA), int(heightB))# 变换后对应坐标位置dst = np.array([[0, 0],[maxWidth - 1, 0],[maxWidth - 1, maxHeight - 1],[0, maxHeight - 1]],dtype=np.float32)# 计算变换矩阵M = cv2.getPerspectiveTransform(rect, dst)print("变换矩阵:\n", M) # 调试输出# 透视变换wraped = cv2.warpPerspective(image, M, (maxWidth, maxHeight))# 返回return wrapeddef resize(image, width=None, height=None, inter=cv2.INTER_AREA):"""修改图片大小:param image: 原图:param width: 宽:param height: 高:param inter: 模式:return: 修改好的图片"""dim = None(h, w) = image.shape[:2]if width is None and height is None:return imageif width is None:r = height / float(h)dim = (int(w * r), height)else:r = width / float(w)dim = (width, int(h * r))resized = cv2.resize(image, dim, interpolation=inter)# 返回return resized

主函数

import argparseimport cv2from pre_process import read_imagefrom pre_process import image_calculate_contoursfrom pre_process import image_transformdef parse_opt():"""设置参数"""parser = argparse.ArgumentParser()parser.add_argument("--image_path", type=str, default="images/receipt2.jpg", help="图片路径")args = parser.parse_args()return argsdef main():"""主函数"""args = parse_opt()# 读取图片image, image_resize, edge, ratio = read_image(image_path=args.image_path, visualize=True)# 计算轮廓approx_array = image_calculate_contours(image_resize, edge, visualize=True)# 图片转换final_result = image_transform(image=image, approx_array=approx_array, ratio=ratio, visualize=True)# 保存最终结果cv2.imwrite("final_result.jpg", final_result)if __name__ == "__main__":main()

输出结果

最终转换结果

原图:

最终结果: