思考题 通过鼠标操作在图像上选取任意4边形区域，通过仿射变换到指定的矩形区域。例如梯形到矩形的变换。
import cv2 import numpy as np # 当鼠标按下时变为True ix1, iy1 = -1, -1 ix2, iy2 = -1, -1 ix3, iy3 = -1, -1 ix4, iy4 = -1, -1 # 创建回调函数 def draw_circle(event, x, y, flags, param): global ix1, iy1, ix2, iy2, ix3, iy3, ix4, iy4 # 当按下左键是返回位置坐标 if event == cv2.EVENT_LBUTTONDOWN: if ix1 == -1: ix1, iy1 = x, y cv2.circle(img, (x, y), 2, (255, 255, 255), -1) elif ix2 == -1: ix2, iy2 = x, y cv2.circle(img, (x, y), 2, (255, 255, 255), -1) elif ix3 == -1: ix3, iy3 = x, y cv2.circle(img, (x, y), 2, (255, 255, 255), -1) elif ix4 == -1: ix4, iy4 = x, y cv2.circle(img, (x, y), 2, (255, 255, 255), -1) img = cv2.imread('shudu.png') cv2.namedWindow('image') cv2.setMouseCallback('image', draw_circle) while 1: cv2.imshow('image', img) k = cv2.waitKey(1) & 0xFF if ix4 != -1: break cv2.imshow('image', img) #cv2.getAffineTransform 会创建一个2x3 的矩阵，最后这个矩阵会被传给函数cv2.warpAffine。 src_point = np.float32([[ix1, iy1], [ix2, iy2], [ix3, iy3], [ix4, iy4]]) dst_point = np.float32([[0, 0], [0, 300], [300, 300], [300, 0]]) h = cv2.getPerspectiveTransform(src_point, dst_point) #通过一一对应的点，计算映射矩阵，进行变换 res = cv2.warpPerspective(img, h, (300, 300)) #cv2.warpPerspective(src, M, dsize[, dst[, flags[, borderMode[, borderValue]]]]) → dst #src：输入图像 # M：变换矩阵 dsize：目标图像shape flags：插值方式，interpolation方法INTER_LINEAR或INTER_NEAREST #borderMode：边界补偿方式，BORDER_CONSTANT or BORDER_REPLICATE #borderValue：边界补偿大小，常值，默认为0 cv2.imshow('output', res) cv2.waitKey(0) cv2.destroyAllWindows()结果如下
