Opencv实现程序-部分12

import cv2 import _pickle as cPickle import numpy as np import gzip # # obj = {"a": 1, "b": 2, "c": 3} # pickle.dump(obj, open("tmp.txt", "wb")) # obj2 = pickle.load(open("tmp.txt", "rb")) # print(obj2) # print(obj2 == obj) def vectorized_result(j): e = np.zeros((10, 1)) e[j] = 1.0 return e # mnist = gzip.open('./data/mnist.pkl.gz', 'rb') training_data, classification_data, test_data = cPickle.load(mnist,encoding='bytes') mnist.close() tr_d, va_d, te_d = training_data, classification_data, test_data # print(tr_d[0].shape) # (50000, 784) 说明每一行的维度是784(图像的像素) # for x in tr_d[0]: 表示每一行 进行转换为 np.reshape(x, (784, 1)) training_inputs = [np.reshape(x, (784, 1)) for x in tr_d[0]] # 转换后的维度： (50000, 784, 1) # print(np.array(training_inputs).shape) # training_results是一个list 转换为np.array后 为：(50000, 10, 1) training_results = [vectorized_result(y) for y in tr_d[1]] # print(type(tr_d[1])) # <class 'numpy.ndarray'> # for y in tr_d[1]: # print(vectorized_result(y)) # print(type(vectorized_result(y))) # print(vectorized_result(y).shape) # print(training_results) # print(np.array(training_results).shape) # training_inputs: (50000, 784, 1) list类型 # training_results： (50000, 10, 1) list类型 training_data = zip(training_inputs, training_results) # print(training_data) # <zip object at 0x0000027701EB4BC8> # print(type(va_d[0])) # <class 'numpy.ndarray'> # print(va_d[0].shape) # (10000, 784) validation_inputs = [np.reshape(x, (784, 1)) for x in va_d[0]] # print(np.array(validation_inputs).shape) # shape: (10000, 784, 1) # print(va_d[1].shape) # (10000,) # validation_inputs： (10000, 784, 1) va_d[1]：(10000,) validation_data = zip(validation_inputs, va_d[1]) # print(validation_data) # <zip object at 0x0000016AB0AC5C48> # print(type(te_d[0])) # <class 'numpy.ndarray'> # print(te_d[0].shape) # (10000, 784) test_inputs = [np.reshape(x, (784, 1)) for x in te_d[0]] # print(np.array(test_inputs).shape) # list类型 shape： (10000, 784, 1) # # print(te_d[1].shape) # np.array() (10000,) # test_data = zip(test_inputs, te_d[1]) # <zip object at 0x0000027701EB4BC8> tr, val, test = training_data, validation_data, test_data # ANN网络的建立 # cv2.ml.ANN_MLP_create函数 ann = cv2.ml.ANN_MLP_create() # 设置网络的层数 ann.setLayerSizes(np.array([784, 20, 10])) # 设置ANN_MLP_RPROP方法 ann.setTrainMethod(cv2.ml.ANN_MLP_RPROP) # cv2.ml.ANN_MLP_SIGMOID_SYM: 设置sigmod为激活函数 ann.setActivationFunction(cv2.ml.ANN_MLP_SIGMOID_SYM) # 设置终止条件ann.setTermCriteria ann.setTermCriteria(( cv2.TERM_CRITERIA_EPS | cv2.TERM_CRITERIA_COUNT, 100, 1 )) samples = 100 counter = 0 # 训练集部分 # 使用for循环来处理类型：zip object at 0x0000027701EB4BC8> for img in tr: if (counter > samples): break if (counter % 10 == 0): print("Trained %d/%d" % (counter, samples)) counter += 1 data, digit = img # # # print(type(data)) # np.array # print(data.shape) # shape: (784, 1) # print(digit.shape) # (10, 1) # 输入： # np.array([data.ravel()], dtype=np.float32) (784,) # np.array([data.ravel()], dtype=np.float32) (10,) ann.train(np.array([data.ravel()], dtype=np.float32), cv2.ml.ROW_SAMPLE, np.array([digit.ravel()], dtype=np.float32)) # print(digit.shape) # (10, 1) # print(digit.ravel().shape) # (10,) # r1 在 r2 内 def inside(r1, r2): x1,y1,w1,h1 = r1 x2,y2,w2,h2 = r2 if (x1 > x2) and (y1 > y2) and (x1+w1 < x2+w2) and (y1+h1< y2 + h2): return True else: return False #### 测试部分实现 font = cv2.FONT_HERSHEY_SIMPLEX path = "./images/numbers.jpg" # 是一个包含多个数字的图片 img = cv2.imread(path, cv2.IMREAD_UNCHANGED) bw = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) bw = cv2.GaussianBlur(bw, (7,7), 0) ret, thbw = cv2.threshold(bw, 135, 255, cv2.THRESH_BINARY_INV) thbw = cv2.erode(thbw, np.ones((2,2), np.uint8), iterations = 2) # cv2.imshow('img',thbw) # cv2.waitKey(0) image, cntrs, hier = cv2.findContours(thbw.copy(), cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE) rectangles = [] # 数字的提取 # 每一个选中的几何区域 for c in cntrs: r = x, y, w, h = cv2.boundingRect(c) a = cv2.contourArea(c) # 整幅img图像大小 b = (img.shape[0] - 3) * (img.shape[1] - 3) # 滤除一些矩形 is_inside = False # rectangles存储的有效矩形 # 打算要不要存储的矩形r for q in rectangles: if inside(r, q): is_inside = True break # if not is_inside: if not a == b: rectangles.append(r) # 将选择的矩形区域变成正方形区域 def wrap_digit(rect): # x, y, w, h x, y, w, h = rect padding = 5 hcenter = x + w/2 vcenter = y + h/2 roi = None if (h > w): w = h x = hcenter - (w/2) else: h = w y = vcenter - (h/2) return (x-padding, y-padding, w+padding, h+padding) # ann: 训练好的网络模型 sample：ROI正方形区域 def predict(ann, sample): resized = sample.copy() rows, cols = resized.shape if (rows != 28 or cols != 28) and rows * cols > 0: resized = cv2.resize(resized, (28, 28), interpolation=cv2.INTER_LINEAR) # ann.predict resized.ravel()--> (784,) return ann.predict(np.array([resized.ravel()], dtype=np.float32)) /* 开发不易，整理也不易，如需要详细的说明文档和程序，以及完整的数据集，训练好的模型，或者进一步开发， 可加作者新联系方式咨询，WX：Q3101759565，QQ：3101759565 */