Handwritten numeral recognition, run your own picture with the saved model

Original blog ：https://blog.csdn.net/X_m_w/article/details/101056156
model training ：https://blog.csdn.net/X1996_/article/details/108883710
Here, we use the training model to predict , Refer to the previous article for model training
Mainly used opencv Process the image into the desired image format , Then predict
Input picture ：
Output ：

import cv2
import numpy as np
from keras import models
#  Inverse grayscale image , Invert the black and white threshold 


def accessPiexl(img):
    height = img.shape[0]
    width = img.shape[1]
    for i in range(height):
       for j in range(width):
           img[i][j] = 255 - img[i][j]
    return img


#  Inverse binary image 
def accessBinary(img, threshold=128):
    img = accessPiexl(img)
    #  Edge expansion , It's OK to 
    kernel = np.ones((3, 3), np.uint8)
    img = cv2.dilate(img, kernel, iterations=1)
    _, img = cv2.threshold(img, threshold, 0, cv2.THRESH_TOZERO)
    return img


#  Use the contour to find the position , No line alignment is required 
def findBorderContours(path, maxArea=50):
    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
    img = accessBinary(img)
    _, contours, _ = cv2.findContours(img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    borders = []
    for contour in contours:
        #  Fit edges into a border 
        x, y, w, h = cv2.boundingRect(contour)
        if w*h > maxArea:
            border = [(x, y), (x+w, y+h)]
            borders.append(border)
    return borders


#  Convert to... According to the border MNIST Format 
def transMNIST(path, borders, size=(28, 28)):
    imgData = np.zeros((len(borders), size[0], size[0], 1), dtype='uint8')
    img = cv2.imread(path, cv2.IMREAD_GRAYSCALE)
    img = accessBinary(img)
    for i, border in enumerate(borders):
        borderImg = img[border[0][1]:border[1][1], border[0][0]:border[1][0]]
        #  Expand pixels according to the maximum edge , Expand the edge border , Fixed value 0 fill , Into a square 
        # top, bottom, left, right： The number of pixels to be expanded up, down, left and right 
        extendPiexl = (max(borderImg.shape) - min(borderImg.shape)) // 2
        #  How much to enlarge depends on the size of your original figure , Put the number in the middle of the picture , The bigger the picture , The expansion is larger 
        targetImg = cv2.copyMakeBorder(borderImg, 30, 30, extendPiexl + 30, extendPiexl + 30, cv2.BORDER_CONSTANT)
        targetImg = cv2.resize(targetImg, size)
        #  Add a dimension  28*28*1, That's how I understand it , amount to reshape Well 
        targetImg = np.expand_dims(targetImg, axis=-1)
        imgData[i] = targetImg
    return imgData


#  Predict handwritten digits 
def predict(modelpath, imgData):
    model = models.load_model(modelpath)
    img = imgData.astype('float32') / 255
    results = model.predict(img)
    result_number = []
    for result in results:
        result_number.append(np.argmax(result))
    return result_number


#  Display results and borders 
def showResults(path, borders, results=None):
    img = cv2.imread(path)
    for i, border in enumerate(borders):
        cv2.rectangle(img, border[0], border[1], (0, 0, 255))
        if results:
            cv2.putText(img, str(results[i]), border[0], cv2.FONT_HERSHEY_COMPLEX, 0.8, (0, 255, 0), 1)
    cv2.imshow('test', img)


path = '11.png'
model = 'model.h5'
borders = findBorderContours(path)
imgData = transMNIST(path, borders)
results = predict(model, imgData)
showResults(path, borders, results)

cv2.waitKey(0)