def pre_img(image): myimage = image.convert('L') # 转换成灰度图 myimage = np.array(myimage) # print(myimage.shape) ret, img1 = cv.threshold(myimage, 100, 255, cv.THRESH_BINARY_INV) # cv.namedWindow('img1',0) # cv.resizeWindow('img1',600,600) # cv.imshow('img1',img1) # print(type(img1)) # print(img1.shape) # print(img1.size) # cv.waitKey(2) kernel1 = np.ones((10, 10), np.uint8) # 做一次膨胀 img2 = cv.dilate(img1, kernel1) # cv.namedWindow('img2', 0) # cv.resizeWindow('img2', 600, 600) # cv.imshow('img2', img2) '剔除小连通域' contours, hierarchy = cv.findContours(img2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_NONE) # print(len(contours),hierarchy) for i in range(len(contours)): area = cv.contourArea(contours[i]) if area < 150: # '设定连通域最小阈值，小于该值被清理' cv.drawContours(img2, [contours[i]], 0, 0, -1) # cv2.imwrite('img2.jpg', img2) # import pdb;pdb.set_trace() img5 = cv.resize(img2, (28, 28)) # cv.namedWindow('img5', 0) # cv.resizeWindow('img5', 600, 600) # cv.imshow('img5', img5) # cv2.imwrite('img5.jpg', img5) # import pdb;pdb.set_trace() return img5 img_pre = pre_img(img) # # cv.imshow('img_pre', img_pre) # 将数据类型由uint8转为float32 img = img_pre.astype(np.float32) # 图片数据归一化 img = img / 255 改成单个代码可以运行

以下是整合后的代码：

import cv2 as cv
import numpy as np

def pre_img(image):
    myimage = image.convert('L') # 转换成灰度图
    myimage = np.array(myimage)
    ret, img1 = cv.threshold(myimage, 100, 255, cv.THRESH_BINARY_INV)
    kernel1 = np.ones((10, 10), np.uint8)
    img2 = cv.dilate(img1, kernel1)
    contours, hierarchy = cv.findContours(img2, cv.RETR_EXTERNAL, cv.CHAIN_APPROX_NONE)
    for i in range(len(contours)):
        area = cv.contourArea(contours[i])
        if area < 150:
            cv.drawContours(img2, [contours[i]], 0, 0, -1)
    img5 = cv.resize(img2, (28, 28))
    return img5

img = cv.imread('example.jpg') # 读入图片
img_pre = pre_img(img)
img = img_pre.astype(np.float32) / 255

cv.imshow('img', img) # 显示图片
cv.waitKey(0) # 等待按键
cv.destroyAllWindows() # 关闭所有窗口

注意需要自行替换掉 `example.jpg` 为你需要处理的图片的路径。