convolution neural network example, for training

Author: kmolLin

cnn_example/IMG_3451.HEIC

@@ -0,0 +1,203 @@
+import cv2
+import numpy as np
+import keras
+from keras.models import Sequential, load_model
+
+
+# 水平方向投影
+def hProject(binary):
+    h, w = binary.shape
+
+    # 水平投影
+    hprojection = np.zeros(binary.shape, dtype=np.uint8)
+
+    # 创建h长度都为0的数组
+    h_h = [0]*h
+    for j in range(h):
+        for i in range(w):
+            if binary[j,i] == 0:
+                h_h[j] += 1
+    # 画出投影图
+    for j in range(h):
+        for i in range(h_h[j]):
+            hprojection[j,i] = 255
+
+    # cv2.imshow('hpro', hprojection)
+
+    return h_h
+
+# 垂直反向投影
+def vProject(binary):
+    h, w = binary.shape
+    # 垂直投影
+    vprojection = np.zeros(binary.shape, dtype=np.uint8)
+
+    # 创建 w 长度都为0的数组
+    w_w = [0]*w
+    for i in range(w):
+        for j in range(h):
+            if binary[j, i ] == 0:
+                w_w[i] += 1
+
+    for i in range(w):
+        for j in range(w_w[i]):
+            vprojection[j,i] = 255
+
+    # cv2.imshow('vpro', vprojection)
+
+    return w_w
+
+
+def load_model_cnn(model, all_imagea):
+    predictt = []
+    for imgf in all_imagea:
+        # img = cv2.imread(f"{i}", 0)
+        img = cv2.resize(imgf, (28, 28))
+        img = 255 - img
+        img = img.astype("float32")
+        img_4 = img - np.amin(img)
+        img_5 = 255 * img_4 / (np.amax(img_4))
+        x_test_img = np.reshape(img_5, (1, 28, 28))
+        x_Test4D = x_test_img.reshape(x_test_img.shape[0], 28, 28, 1).astype('float32')
+        x_Test4D_normalize = (x_Test4D / np.amax(x_test_img))
+        prediction = model.predict_classes(x_Test4D_normalize)
+        predictt.append(prediction)
+        # .append(prediction)
+    return predictt
+
+
+def load_model_cnn_unit(model, image):
+    img = cv2.resize(image, (28, 28))
+    img = 255 - img
+    img = img.astype("float32")
+    img_4 = img - np.amin(img)
+    img_5 = 255 * img_4 / (np.amax(img_4))
+    x_test_img = np.reshape(img_5, (1, 28, 28))
+    x_Test4D = x_test_img.reshape(x_test_img.shape[0], 28, 28, 1).astype('float32')
+    x_Test4D_normalize = (x_Test4D / np.amax(x_test_img))
+    prediction = model.predict_classes(x_Test4D_normalize)
+    return prediction
+
+def trackChaned(x):
+  pass
+
+if __name__ == '__main__':
+    all_image = []
+    video = cv2.VideoCapture(1)
+    model = load_model('my_model.h5')
+    model.load_weights('my_model_weights.h5')
+
+    # video.set(cv2.CAP_PROP_AUTOFOCUS, 0)
+    # frame_width = int(video.get(cv2.CAP_PROP_FRAME_WIDTH))
+    # frame_height = int(video.get(cv2.CAP_PROP_FRAME_HEIGHT))
+    # avi = cv2.VideoWriter_fourcc(*'MP4V')
+    # out = cv2.VideoWriter("test123.mov", avi, 25, (1200, 600))
+    lower_blue = np.array([78, 43, 46])
+    upper_blue = np.array([110, 255, 255])
+    
+    cv2.namedWindow('Mask')
+    cv2.createTrackbar("Min", "Mask", 0, 255, trackChaned)
+    if not video.isOpened():
+        print("Could not open video")
+        sys.exit()
+    ok, frame = video.read()
+    if not ok:
+        print('Cannot read video file')
+        sys.exit()
+
+    tmp = None
+    # cv2.namedWindow("Tracking", cv2.WINDOW_NORMAL)
+    while True:
+        all_imagea = []
+        # Read a new frame
+        ok, frame = video.read()
+        gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+        huh = cv2.getTrackbarPos("Min", "Mask")
+        ret, th = cv2.threshold(gray, huh, 255, 0)
+        cv2.imshow('originla', frame)
+        cv2.imshow('Mask', th)
+        
+        # Exit if ESC pressed
+        k = cv2.waitKey(100) & 0xff
+        if k == 27:
+            cv2.destroyAllWindows()
+            cv2.waitKey(1)
+            break
+        if k == ord('p'):
+            print("press the key == p")
+            # framee = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+            # ret, th = cv2.threshold(framee, 127, 255, 0)
+            h, w = gray.shape
+            h_h = hProject(th)
+
+            start = 0
+            h_start, h_end = [], []
+            position = []
+
+            # 根据水平投影获取垂直分割
+            for i in range(len(h_h)):
+                if h_h[i] > 0 and start == 0:
+                    h_start.append(i)
+                    start = 1
+                if h_h[i] == 0 and start == 1:
+                    h_end.append(i)
+                    start = 0
+
+            for i in range(len(h_start)):
+                cropImg = th[h_start[i]:h_end[i], 0:w]
+                if i == 0:
+                    pass
+                    # cv2.imshow('cropimg', cropImg)
+                    # cv2.imwrite('words_cropimg.jpg', cropImg)
+                w_w = vProject(cropImg)
+
+                wstart , wend, w_start, w_end = 0, 0, 0, 0
+                for j in range(len(w_w)):
+                    if w_w[j] > 0 and wstart == 0:
+                        w_start = j
+                        wstart = 1
+                        wend = 0
+                    if w_w[j] ==0 and wstart == 1:
+                        w_end = j
+                        wstart = 0
+                        wend = 1
+
+                    # 当确认了起点和终点之后保存坐标   
+                    if wend == 1:
+                        position.append([w_start, h_start[i], w_end, h_end[i]])
+                        wend = 0
+
+                # 确定分割位置
+                for i, p in enumerate(position):
+                    height = abs(p[1] - p[3])
+                    weidgh = abs(p[0] - p[2])
+                    y = height if height > weidgh else weidgh
+                    x = height if height > weidgh else weidgh
+                    # print(p)
+                    # int((y - height) / 2)
+                    center_point = (int((p[0] + p[2]) / 2), int((p[1] + p[3]) / 2))
+                    print(f"center point {center_point}")
+                    # print(int((center_point[0] - y / 2)), int((center_point[0] + y / 2)))
+                    # print(int((center_point[0] - y / 2)), int((center_point[1] + x / 2)))
+                    imgg = th[int((center_point[1] - y / 2)):int((center_point[1] + y / 2)),
+                              int((center_point[0] - x / 2)):int((center_point[0] + x / 2))]
+                    # print(int((y - height) / 2), int(y - (y - height) / 2))
+                    # imgg = th[int((p[0] - height) / 2): int(p[0] - (p[0] - height) / 2),
+                    #           int((p[1] - weidgh) / 2): int(p[1] - (p[1] - weidgh) / 2)]
+                    imgg = cv2.resize(imgg, (28, 28))
+                    # cv2.imwrite(f"{i}.jpg", imgg)
+                    cv2.imshow(f"{i}", imgg)
+                    # all_image.append(imgg)
+                    predic = load_model_cnn_unit(model, imgg)
+                    print(predic[0])
+                    cv2.putText(frame, f"{predic[0]}", (center_point[0], center_point[1] - 50),
+                                cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 1, cv2.LINE_AA)
+                    cv2.rectangle(frame, (int((center_point[0] - x / 2)), int((center_point[1] - y / 2))),
+                                  (int((center_point[0] + x / 2)), int((center_point[1] + y / 2))), (0, 0, 255), 2)
+            # print(load_model_cnn(model, all_image))
+            # cv2.imshow("th", th)
+            cv2.imshow("test", frame)
+
+    video.release()
+    cv2.destroyAllWindows() 
+

cnn_example/character_cutting.py

@@ -0,0 +1,34 @@
+# -*- coding: utf-8 -*-
+import os
+from keras.models import Sequential, load_model
+from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D
+from keras.utils import np_utils, plot_model
+from keras.datasets import mnist
+from keras.preprocessing import image
+import matplotlib.pyplot as plt
+import numpy as np
+import cv2
+# import pandas as pd
+
+# Load Model
+model = load_model('my_model.h5')
+model.load_weights('my_model_weights.h5')
+Path = ["3.jpg", "6.jpg", "10.jpg"]
+
+print(model.input_shape)
+print(model.summary())
+# img = image.load_img("0.jpg", 0)
+for i in Path:
+    
+    img = cv2.imread(f"{i}", 0)
+    img = cv2.resize(img, (28, 28))
+    img = 255 - img
+    img = img.astype("float32")
+    img_4 = img - np.amin(img)
+    img_5 = 255 * img_4 / (np.amax(img_4))
+    x_test_img = np.reshape(img_5, (1, 28, 28))
+    x_Test4D = x_test_img.reshape(x_test_img.shape[0], 28, 28, 1).astype('float32')
+    x_Test4D_normalize = (x_Test4D / np.amax(x_test_img))
+    prediction = model.predict_classes(x_Test4D_normalize)
+
+    print(prediction)