一、自动获取图像顶点变换(获取图像轮廓顶点矫正)
图像旋转校正思路如下
1、以灰度图读入
2、腐蚀膨胀,闭合等操作
3、二值化图像
4、获取图像顶点
5、透视矫正
#(基于透视的图像矫正)import cv2import mathimport numpy as npdef Img_Outline(input_dir):original_img = cv2.imread(input_dir)gray_img = cv2.cvtColor(original_img, cv2.COLOR_BGR2GRAY)blurred = cv2.GaussianBlur(gray_img, (9, 9), 0) # 高斯模糊去噪(设定卷积核大小影响效果)_, RedThresh = cv2.threshold(blurred, 165, 255, cv2.THRESH_BINARY) # 设定阈值165(阈值影响开闭运算效果)kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (5, 5))# 定义矩形结构元素closed = cv2.morphologyEx(RedThresh, cv2.MORPH_CLOSE, kernel) # 闭运算(链接块)opened = cv2.morphologyEx(closed, cv2.MORPH_OPEN, kernel) # 开运算(去噪点)return original_img, gray_img, RedThresh, closed, openeddef findContours_img(original_img, opened):image, contours, hierarchy = cv2.findContours(opened, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)c = sorted(contours, key=cv2.contourArea, reverse=True)[1] # 计算最大轮廓的旋转包围盒rect = cv2.minAreaRect(c)# 获取包围盒(中心点,宽高,旋转角度)box = np.int0(cv2.boxPoints(rect)) # boxdraw_img = cv2.drawContours(original_img.copy(), [box], -1, (0, 0, 255), 3)print("box[0]:", box[0])print("box[1]:", box[1])print("box[2]:", box[2])print("box[3]:", box[3])return box,draw_imgdef Perspective_transform(box,original_img):# 获取画框宽高(x=orignal_W,y=orignal_H)orignal_W = math.ceil(np.sqrt((box[3][1] - box[2][1])**2 + (box[3][0] - box[2][0])**2))orignal_H= math.ceil(np.sqrt((box[3][1] - box[0][1])**2 + (box[3][0] - box[0][0])**2))# 原图中的四个顶点,与变换矩阵pts1 = np.float32([box[0], box[1], box[2], box[3]])pts2 = np.float32([[int(orignal_W+1),int(orignal_H+1)], [0, int(orignal_H+1)], [0, 0], [int(orignal_W+1), 0]])# 生成透视变换矩阵;进行透视变换M = cv2.getPerspectiveTransform(pts1, pts2)result_img = cv2.warpPerspective(original_img, M, (int(orignal_W+3),int(orignal_H+1)))return result_imgif __name__=="__main__":input_dir = "../staticimg/oldimg_04.jpg"original_img, gray_img, RedThresh, closed, opened = Img_Outline(input_dir)box, draw_img = findContours_img(original_img,opened)result_img = Perspective_transform(box,original_img)cv2.imshow("original", original_img)cv2.imshow("gray", gray_img)cv2.imshow("closed", closed)cv2.imshow("opened", opened)cv2.imshow("draw_img", draw_img)cv2.imshow("result_img", result_img)cv2.waitKey(0)cv2.destroyAllWindows()
直接变换
1、获取图像四个顶点
2、形成变换矩阵
3、透视变换
import cv2import numpy as npimport matplotlib.pyplot as pltimg = cv2.imread('original_img.jpg')H_rows, W_cols= img.shape[:2]print(H_rows, W_cols)# 原图中书本的四个角点(左上、右上、左下、右下),与变换后矩阵位置pts1 = np.float32([[161, 80], [449, 12], [1, 430], [480, 394]])pts2 = np.float32([[0, 0],[W_cols,0],[0, H_rows],[H_rows,W_cols],])# 生成透视变换矩阵;进行透视变换M = cv2.getPerspectiveTransform(pts1, pts2)dst = cv2.warpPerspective(img, M, (500,470))"""注释代码同效# img[:, :, ::-1]是将BGR转化为RGB# plt.subplot(121), plt.imshow(img[:, :, ::-1]), plt.title('input')# plt.subplot(122), plt.imshow(dst[:, :, ::-1]), plt.title('output')# plt.show"""cv2.imshow("original_img",img)cv2.imshow("result",dst)cv2.waitKey(0)cv2.destroyAllWindows()
两次透视变换
def get_warp_perspective(img, width, height, array_points, array_points_get, array_points_warp):middle_len = 268# rows, cols = img.shape[:2]# D_value1 = (middle_len - array_points_get[0][1])*2+((middle_len - array_points_get[0][1])//3)# D_value2 = (middle_len - array_points_get[1][1])*2+((middle_len - array_points_get[1][1])//3)D_value1 = 0D_value2 = 0# 原图中的四个角点# pts1 = np.float32([[0, 249],[512, 253],[0, 512], [512, 512]])#重要的测试1和2pts1 = np.float32(array_points_get)#重要的测试1和2# pts2 = np.float32([[0, middle_len], [width, middle_len], [0, height], [width, height]])#重要的测试1和2# pts2 = np.float32([[0, middle_len],[0, height] , [width, height],[width, middle_len]])#重要的测试1和2pts2 = np.float32([[0, 0],[0, middle_len] , [width, middle_len],[width, 0]])#重要的测试1和2# 生成透视变换矩阵M = cv2.getPerspectiveTransform(pts1, pts2)# 进行透视变换dst = cv2.warpPerspective(img, M, (width, height))# # 保存图片,仅用于测试img_path = './cut_labels/cut_image_one.jpg'cv2.imwrite(img_path, dst)return warp_perspective(dst, width, height,array_points,array_points_warp,middle_len, D_value1, D_value2)def warp_perspective(dst, width, height,array_points,array_points_warp,middle_len, D_value1, D_value2):# new_img_path = img_path# img = cv2.imread(new_img_path)# 原图的保存地址# rows, cols = img.shape[:2]# 原图中的四个角点# pts3 = np.float32([[0, 268], [0, 44], [512,35], [512, 268]])#重要测试1# pts3 = np.float32([[0, middle_len], [0, D_value1], [512,D_value2], [512, middle_len]])#重要测试1pts3 = np.float32([[0, 0], [0, height], [width, height], [width, 0]])# pts3 = np.float32([[0, middle_len], [0, D_value1], [512,D_value2], [512, middle_len]])#重要测试1# pts3 = np.float32([[0, 512], [0, array_points[1][1]], [512,512], [512, middle_len]])#重要测试1# 变换后的四个角点pts4 = np.float32([[0, 0], [0, height-D_value1], [width, height-D_value2], [width, 0]])#重要测试1# pts4 = np.float32([[0, 268], [0, 0], [512, 0], [512, 268]])#重要测试1# 生成透视变换矩阵M = cv2.getPerspectiveTransform(pts3, pts4)# 进行透视变换dst_img = cv2.warpPerspective(dst, M, (width, height))# #保存最终图片,仅用于测试print("++++++++++++++++")final_img_path = './cut_labels/cut_image_two.jpg'cv2.imwrite(final_img_path, dst_img)# 进行透视变换return cv2.warpPerspective(dst_img, M, (width, height))# return output_warp_perspective(img, width, height, array_points, array_points_get, array_points_warp)if __name__ == "__main__":# 透视转换img = cv2.imread('../staticimg/oldimg_04.jpg')dst = get_warp_perspective(img, 512, 512, array_points=[[395.2, 75.0], [342, 517], [1000, 502], [900, 75]])cv2.imwrite('aaa2.jpg', dst)cv2.imshow('title', dst)cv2.waitKey(0)imgrectificate = imgRectificate(img, width, height, array_points)imgrectificate.warp_perspective()