为什么有些验证码看起来很容易但是没人做自动识别的？ 第1页

       cv2.cvtColor(img, cv2.COLOR_GRAY2RGB) ret, img = cv2.threshold(img, 180, 255, cv2.THRESH_BINARY)     

       def img_salt_filter(img):     """     去除椒盐类干扰     :param img:     :return:     """     newIMG = cv2.medianBlur(img, 5)     return newIMG  def img_open_filter(img):     """     图像开运算，去除噪点类干扰以及补全验证码缺口     :param img:     :return:     """     kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (2, 2))     kernel2 = cv2.getStructuringElement(cv2.MORPH_CROSS, (2, 2))     newImg = cv2.morphologyEx(img, cv2.MORPH_OPEN, kernel)     newImg = cv2.dilate(newImg, kernel2)     return newImg   def img_close_filter(img):     """     图像闭运算，去除干扰     :param img:     :return:     """     kernel = cv2.getStructuringElement(cv2.MORPH_CROSS, (2, 2))     kernel2 = cv2.getStructuringElement(cv2.MORPH_CROSS, (1, 1))     closed = cv2.morphologyEx(img, cv2.MORPH_CLOSE, kernel)     newImg = cv2.erode(closed, kernel2, iterations=1)     return newImg  def neighbo_filter(img)     """     邻域扫描运算，去除干扰     :param img:     :return:     """     这个就是扫描一个像素点附近是否存在其他点，很简单，就不贴出来了     

       def img_mser(img):     """     图像切割算法     :param img:     :return:     """     # 图像预处理     img = img_resize(img)     box_content = list()     box_mser = list()      mser = cv2.MSER_create(_min_area=5, _max_area=300)     img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)     ret, img = cv2.threshold(img, 200, 255, cv2.THRESH_BINARY)     # img_show(img)     # try:      # 连通域划分及去重     regions, boxes = mser.detectRegions(img)     print len(boxes)     boxes = boxes.tolist()     for i in boxes:         if i not in box_content:             box_content.append(i)       # 使用欧氏距离算法去除近似连通域     for j in box_content:         if box_mser:             for k in xrange(len(box_mser)):                 if Euclidean(j, box_mser[k]) > 4 and k == len(box_mser) - 1:                     box_mser.append(j)         else:             box_mser.append(j)      # 去除多余连通域     box_mser_len = len(box_mser)     if box_mser_len != 4:         if box_mser_len > 4:             for i in box_mser:                 area = list()                 area.append(i[2] * i[3])             for j in xrange(box_mser_len - 4):                 pos = area.index(min(area)) - 1                 box_mser.remove(box_mser[pos])         else:             print u'切割出现问题，请改变设置值'      # print box_mser     cut_img = list()     box_mser.sort()     for z in xrange(len(box_mser)):         k = box_mser[z]         slice_img = img[k[1]-1:k[1] + k[3]+3, k[0]-2:k[0] + k[2] + 2]         # img_show(slice_img)         cut_img.append(cv2.resize(slice_img, (30, 40)))     return cut_img  def img_resize(img):     img = cv2.resize(img, (50, 20))     return img  def Euclidean(vec1, vec2):     """     Euclidean_Distance,欧式距离     :param vec1:     :param vec2:     :return:     """     npvec1, npvec2 = np.array(vec1), np.array(vec2)     return math.sqrt(((npvec1 - npvec2) ** 2).sum())     

       model = Sequential()     inputShape = (height, width, depth)     # 第一层卷积     model.add(         convolutional.Conv2D(filters=32, kernel_size=(5, 5), input_shape=inputShape, data_format="channels_last")     )      model.add(Activation('relu'))      model.add(MaxPooling2D(         pool_size=(2, 2),         strides=(2, 2),         padding='same',     ))      # 第二层卷积     model.add(         convolutional.Conv2D(filters=128, kernel_size=(3, 3), input_shape=inputShape, data_format="channels_last")     )     model.add(Activation('relu'))     model.add(MaxPooling2D(         pool_size=(2, 2),         padding='same',     ))      # 第三层卷积     model.add(         convolutional.Conv2D(filters=128, kernel_size=(3, 3), input_shape=inputShape, data_format="channels_last")     )     model.add(Activation('relu'))     model.add(MaxPooling2D(         pool_size=(2, 2),         padding='same',     ))      # # 第四层卷积     # model.add(     #     convolutional.Conv2D(filters=128, kernel_size=(2, 2), input_shape=inputShape, data_format="channels_last")     # )     # model.add(Activation('relu'))     # model.add(MaxPooling2D(     #     pool_size=(2, 2),     #     padding='same',     # ))      # 三维向量一维化     model.add(Flatten())     '''         第一层全连接层     '''      model.add(Dense(1024))     model.add(Activation('relu'))     model.add(Dropout(0.2))      '''         第二层全连接层     '''     model.add(Dense(CLASS_NUM))     model.add(Activation('softmax'))      adam = Adam(1e-4)     model.compile(         optimizer=adam,         loss='categorical_crossentropy',         metrics=['accuracy']     )     return model