#!/usr/bin/env python3 # -*- coding: utf-8 -*- """ Created on Thu Sep 9 23:11:51 2020 table line detect @author: chineseocr """ import copy import tensorflow as tf import tensorflow.keras as K from tensorflow.keras.models import Model from tensorflow.keras.layers import Input, concatenate, Conv2D, MaxPooling2D, BatchNormalization, UpSampling2D from tensorflow.keras.layers import LeakyReLU from config import tableModeLinePath from utils import letterbox_image, get_table_line, adjust_lines, line_to_line, draw_boxes import numpy as np import cv2 import time from utils import draw_lines def dice_coef(y_true, y_pred, smooth=1e-5): y_true_f = K.flatten(y_true) y_pred_f = K.flatten(y_pred) intersection = K.sum(y_true_f * y_pred_f) return (2. * intersection + smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) + smooth) def dice_coef_loss(): def dice_coef_loss_fixed(y_true, y_pred): return -dice_coef(y_true, y_pred) return dice_coef_loss_fixed def zero_loss(): def zero_loss_fixed(y_true, y_pred): return tf.zeros_like(y_pred) return zero_loss_fixed def focal_loss(gamma=3., alpha=.5): # 3 0.85 2000e acc-0.6 p-0.99 r-0.99 val_acc-0.56 val_p-0.86 val_r-0.95 # 2 0.85 double_gpu acc- # 3 0.25 gpu 50e acc-0.5 p-0.99 r-0.99 val_acc-0.45 val_p-0.96 val_r-0.88 # 2 0.25 gpu acc- # 3 0.5 double_gpu acc-0.6 p-0.99 r-0.99 val_acc-0.60 val_p-0.93 val_r-0.93 def focal_loss_fixed(y_true, y_pred): pt_1 = tf.where(tf.equal(y_true, 1), y_pred, tf.ones_like(y_pred)) pt_0 = tf.where(tf.equal(y_true, 0), y_pred, tf.zeros_like(y_pred)) return -K.backend.sum(alpha * K.backend.pow(1. - pt_1, gamma) * K.backend.log(K.backend.epsilon()+pt_1))-K.backend.sum((1-alpha) * K.backend.pow( pt_0, gamma) * K.backend.log(1. - pt_0 + K.backend.epsilon())) return focal_loss_fixed def table_net_large(input_shape=(1152, 896, 3), num_classes=1): inputs = Input(shape=input_shape) # 512 use_bias = False down0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(inputs) down0a = BatchNormalization()(down0a) down0a = LeakyReLU(alpha=0.1)(down0a) down0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(down0a) down0a = BatchNormalization()(down0a) down0a = LeakyReLU(alpha=0.1)(down0a) down0a_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0a) # 256 down0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(down0a_pool) down0 = BatchNormalization()(down0) down0 = LeakyReLU(alpha=0.1)(down0) down0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(down0) down0 = BatchNormalization()(down0) down0 = LeakyReLU(alpha=0.1)(down0) down0_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0) # 128 down1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(down0_pool) down1 = BatchNormalization()(down1) down1 = LeakyReLU(alpha=0.1)(down1) down1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(down1) down1 = BatchNormalization()(down1) down1 = LeakyReLU(alpha=0.1)(down1) down1_pool = MaxPooling2D((2, 2), strides=(2, 2))(down1) # 64 down2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(down1_pool) down2 = BatchNormalization()(down2) down2 = LeakyReLU(alpha=0.1)(down2) down2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(down2) down2 = BatchNormalization()(down2) down2 = LeakyReLU(alpha=0.1)(down2) down2_pool = MaxPooling2D((2, 2), strides=(2, 2))(down2) # 32 down3 = Conv2D(256, (3, 3), padding='same', use_bias=use_bias)(down2_pool) down3 = BatchNormalization()(down3) down3 = LeakyReLU(alpha=0.1)(down3) down3 = Conv2D(256, (3, 3), padding='same', use_bias=use_bias)(down3) down3 = BatchNormalization()(down3) down3 = LeakyReLU(alpha=0.1)(down3) down3_pool = MaxPooling2D((2, 2), strides=(2, 2))(down3) # 16 down4 = Conv2D(512, (3, 3), padding='same', use_bias=use_bias)(down3_pool) down4 = BatchNormalization()(down4) down4 = LeakyReLU(alpha=0.1)(down4) down4 = Conv2D(512, (3, 3), padding='same', use_bias=use_bias)(down4) down4 = BatchNormalization()(down4) down4 = LeakyReLU(alpha=0.1)(down4) down4_pool = MaxPooling2D((2, 2), strides=(2, 2))(down4) # 8 center = Conv2D(1024, (3, 3), padding='same', use_bias=use_bias)(down4_pool) center = BatchNormalization()(center) center = LeakyReLU(alpha=0.1)(center) center = Conv2D(1024, (3, 3), padding='same', use_bias=use_bias)(center) center = BatchNormalization()(center) center = LeakyReLU(alpha=0.1)(center) # center up4 = UpSampling2D((2, 2))(center) up4 = concatenate([down4, up4], axis=3) up4 = Conv2D(512, (3, 3), padding='same', use_bias=use_bias)(up4) up4 = BatchNormalization()(up4) up4 = LeakyReLU(alpha=0.1)(up4) up4 = Conv2D(512, (3, 3), padding='same', use_bias=use_bias)(up4) up4 = BatchNormalization()(up4) up4 = LeakyReLU(alpha=0.1)(up4) up4 = Conv2D(512, (3, 3), padding='same', use_bias=use_bias)(up4) up4 = BatchNormalization()(up4) up4 = LeakyReLU(alpha=0.1)(up4) # 16 up3 = UpSampling2D((2, 2))(up4) up3 = concatenate([down3, up3], axis=3) up3 = Conv2D(256, (3, 3), padding='same', use_bias=use_bias)(up3) up3 = BatchNormalization()(up3) up3 = LeakyReLU(alpha=0.1)(up3) up3 = Conv2D(256, (3, 3), padding='same', use_bias=use_bias)(up3) up3 = BatchNormalization()(up3) up3 = LeakyReLU(alpha=0.1)(up3) up3 = Conv2D(256, (3, 3), padding='same', use_bias=use_bias)(up3) up3 = BatchNormalization()(up3) up3 = LeakyReLU(alpha=0.1)(up3) # 32 up2 = UpSampling2D((2, 2))(up3) up2 = concatenate([down2, up2], axis=3) up2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(up2) up2 = BatchNormalization()(up2) up2 = LeakyReLU(alpha=0.1)(up2) up2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(up2) up2 = BatchNormalization()(up2) up2 = LeakyReLU(alpha=0.1)(up2) up2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(up2) up2 = BatchNormalization()(up2) up2 = LeakyReLU(alpha=0.1)(up2) # 64 up1 = UpSampling2D((2, 2))(up2) up1 = concatenate([down1, up1], axis=3) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) # 128 up0 = UpSampling2D((2, 2))(up1) up0 = concatenate([down0, up0], axis=3) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) # 256 up0a = UpSampling2D((2, 2))(up0) up0a = concatenate([down0a, up0a], axis=3) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) # 512 classify = Conv2D(num_classes, (1, 1), activation='sigmoid')(up0a) model = Model(inputs=inputs, outputs=classify) return model def table_net(input_shape=(1152, 896, 3), num_classes=1): inputs = Input(shape=input_shape) # 512 use_bias = False down0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(inputs) down0a = BatchNormalization()(down0a) down0a = LeakyReLU(alpha=0.1)(down0a) down0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(down0a) down0a = BatchNormalization()(down0a) down0a = LeakyReLU(alpha=0.1)(down0a) down0a_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0a) # 256 down0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(down0a_pool) down0 = BatchNormalization()(down0) down0 = LeakyReLU(alpha=0.1)(down0) down0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(down0) down0 = BatchNormalization()(down0) down0 = LeakyReLU(alpha=0.1)(down0) down0_pool = MaxPooling2D((2, 2), strides=(2, 2))(down0) # 128 down1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(down0_pool) down1 = BatchNormalization()(down1) down1 = LeakyReLU(alpha=0.1)(down1) down1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(down1) down1 = BatchNormalization()(down1) down1 = LeakyReLU(alpha=0.1)(down1) down1_pool = MaxPooling2D((2, 2), strides=(2, 2))(down1) # 64 down2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(down1_pool) down2 = BatchNormalization()(down2) down2 = LeakyReLU(alpha=0.1)(down2) down2 = Conv2D(128, (3, 3), padding='same', use_bias=use_bias)(down2) down2 = BatchNormalization()(down2) down2 = LeakyReLU(alpha=0.1)(down2) down2_pool = MaxPooling2D((2, 2), strides=(2, 2))(down2) # 32 up1 = UpSampling2D((2, 2))(down2) up1 = concatenate([down1, up1], axis=3) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) up1 = Conv2D(64, (3, 3), padding='same', use_bias=use_bias)(up1) up1 = BatchNormalization()(up1) up1 = LeakyReLU(alpha=0.1)(up1) # 128 up0 = UpSampling2D((2, 2))(up1) up0 = concatenate([down0, up0], axis=3) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) up0 = Conv2D(32, (3, 3), padding='same', use_bias=use_bias)(up0) up0 = BatchNormalization()(up0) up0 = LeakyReLU(alpha=0.1)(up0) # 256 up0a = UpSampling2D((2, 2))(up0) up0a = concatenate([down0a, up0a], axis=3) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) up0a = Conv2D(16, (3, 3), padding='same', use_bias=use_bias)(up0a) up0a = BatchNormalization()(up0a) up0a = LeakyReLU(alpha=0.1)(up0a) # 512 classify = Conv2D(num_classes, (1, 1), activation='sigmoid')(up0a) model = Model(inputs=inputs, outputs=classify) return model model = table_net((None, None, 3), 2) def table_line(img, model, size=(512, 1024), hprob=0.5, vprob=0.5, row=50, col=30, alph=15): sizew, sizeh = size # [..., ::-1] 最后一维内部反向输出 # inputBlob, fx, fy = letterbox_image(img[..., ::-1], (sizew, sizeh)) # pred = model.predict(np.array([np.array(inputBlob)])) # pred = model.predict(np.array([np.array(inputBlob)/255.0])) img_new = cv2.resize(img, (sizew, sizeh), interpolation=cv2.INTER_AREA) pred = model.predict(np.array([img_new])) pred = pred[0] vpred = pred[..., 1] > vprob # 横线 hpred = pred[..., 0] > hprob # 竖线 vpred = vpred.astype(int) hpred = hpred.astype(int) # print("vpred shape", vpred) # print("hpred shape", hpred) colboxes = get_table_line(vpred, axis=1, lineW=col) rowboxes = get_table_line(hpred, axis=0, lineW=row) # if len(rowboxes) > 0: # rowboxes = np.array(rowboxes) # rowboxes[:, [0, 2]] = rowboxes[:, [0, 2]]/fx # rowboxes[:, [1, 3]] = rowboxes[:, [1, 3]]/fy # rowboxes = rowboxes.tolist() # if len(colboxes) > 0: # colboxes = np.array(colboxes) # colboxes[:, [0, 2]] = colboxes[:, [0, 2]]/fx # colboxes[:, [1, 3]] = colboxes[:, [1, 3]]/fy # colboxes = colboxes.tolist() # nrow = len(rowboxes) # ncol = len(colboxes) # for i in range(nrow): # for j in range(ncol): # rowboxes[i] = line_to_line(rowboxes[i], colboxes[j], 10) # colboxes[j] = line_to_line(colboxes[j], rowboxes[i], 10) # 删掉贴着边框的line temp_list = [] threshold = 5 for line in rowboxes: if line[1]-0 <= threshold or size[1]-line[1] <= threshold: continue temp_list.append(line) rowboxes = temp_list temp_list = [] for line in colboxes: if line[0]-0 <= threshold or size[0]-line[0] <= threshold: continue temp_list.append(line) colboxes = temp_list return rowboxes, colboxes, img_new def get_outline(points, image_np): # 取出x, y的最大值最小值 x_min = points[0][0] x_max = points[-1][0] points.sort(key=lambda x: (x[1], x[0])) y_min = points[0][1] y_max = points[-1][1] # 创建空图 # outline_img = np.zeros(image_size, np.uint8) outline_img = np.copy(image_np) cv2.rectangle(outline_img, (x_min-5, y_min-5), (x_max+5, y_max+5), (0, 0, 0), 2) # cv2.imshow("outline_img", outline_img) # cv2.waitKey(0) return outline_img def get_split_line(points, col_lines, image_np): # print("get_split_line", image_np.shape) points.sort(key=lambda x: (x[1], x[0])) # 遍历y坐标，并判断y坐标与上一个y坐标是否存在连接线 i = 0 split_line_y = [] for point in points: # 从已分开的线下面开始判断 if split_line_y: if point[1] <= split_line_y[-1] + 5: last_y = point[1] continue if last_y <= split_line_y[-1] + 5: last_y = point[1] continue if i == 0: last_y = point[1] i += 1 continue current_line = (last_y, point[1]) split_flag = 1 for col in col_lines: # 只要找到一条col包含就不是分割线 if current_line[0] >= col[1]-3 and current_line[1] <= col[3]+3: split_flag = 0 # print("img", img.shape) # print("col", col) # print("current_line", current_line) break if split_flag: split_line_y.append(current_line[0]+5) split_line_y.append(current_line[1]-5) last_y = point[1] # 加上收尾分割线 points.sort(key=lambda x: (x[1], x[0])) y_min = points[0][1] y_max = points[-1][1] # print("加上收尾分割线", y_min, y_max) if y_min-5 < 0: split_line_y.append(0) else: split_line_y.append(y_min-5) if y_max+5 > image_np.shape[0]: split_line_y.append(image_np.shape[0]) else: split_line_y.append(y_max+5) split_line_y = list(set(split_line_y)) # 剔除两条相隔太近分割线 temp_split_line_y = [] split_line_y.sort(key=lambda x: x) last_y = -20 for y in split_line_y: # print(y) if y - last_y >= 20: # print(y, last_y) temp_split_line_y.append(y) last_y = y split_line_y = temp_split_line_y # print("split_line_y", split_line_y) # 生成分割线 split_line = [] last_y = 0 for y in split_line_y: # if y - last_y <= 15: # continue split_line.append([(0, y), (image_np.shape[1], y)]) last_y = y split_line.append([(0, 0), (image_np.shape[1], 0)]) split_line.append([(0, image_np.shape[0]), (image_np.shape[1], image_np.shape[0])]) split_line.sort(key=lambda x: x[0][1]) # print("split_line", split_line) # 画图画线 # split_line_img = np.copy(image_np) # for y in split_line_y: # cv2.line(split_line_img, (0, y), (image_np.shape[1], y), (0, 0, 0), 1) # cv2.imshow("split_line_img", split_line_img) # cv2.waitKey(0) return split_line, split_line_y def get_points(row_lines, col_lines, image_size): # 创建空图 row_img = np.zeros(image_size, np.uint8) col_img = np.zeros(image_size, np.uint8) # 画线 thresh = 3 for row in row_lines: cv2.line(row_img, (int(row[0]-thresh), int(row[1])), (int(row[2]+thresh), int(row[3])), (255, 255, 255), 1) for col in col_lines: cv2.line(col_img, (int(col[0]), int(col[1]-thresh)), (int(col[2]), int(col[3]+thresh)), (255, 255, 255), 1) # 求出交点 point_img = np.bitwise_and(row_img, col_img) # cv2.imshow("point_img", np.bitwise_not(point_img)) # cv2.waitKey(0) # 识别黑白图中的白色交叉点，将横纵坐标取出 ys, xs = np.where(point_img > 0) points = [] for i in range(len(xs)): points.append((xs[i], ys[i])) points.sort(key=lambda x: (x[0], x[1])) return points def get_minAreaRect(image): gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) gray = cv2.bitwise_not(gray) thresh = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU)[1] coords = np.column_stack(np.where(thresh > 0)) return cv2.minAreaRect(coords) def get_contours(image): gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) ret, binary = cv2.threshold(gray, 127, 255, cv2.THRESH_BINARY) contours, hierarchy = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE) cv2.drawContours(image, contours, -1, (0, 0, 255), 3) cv2.imshow("get contours", image) cv2.waitKey(0) def fix_outline(image, row_lines, col_lines, points, split_y): # 分割线纵坐标 # print("len(split_y)", split_y) if len(split_y) < 2: # split_y = [2000., 2000., 2000., 2000.] return [], [], [] elif len(split_y) == 2: split_y = [2000., 2000., 2000., 2000.] split_y.sort(key=lambda x: x) split_y = split_y[1:-1] new_split_y = [] for i in range(1, len(split_y), 2): new_split_y.append(int((split_y[i]+split_y[i-1])/2)) # print("len(new_split_y)", len(new_split_y)) # 预测线根据分割线纵坐标分为多个分割区域 row_lines.sort(key=lambda x: (x[3], x[2], x[1], x[0])) col_lines.sort(key=lambda x: (x[3], x[2], x[1], x[0])) row_count = 0 col_count = 0 split_row_list = [] split_col_list = [] for y in new_split_y: row_lines = row_lines[row_count:] col_lines = col_lines[col_count:] row_count = 0 col_count = 0 for row in row_lines: if row[3] <= y: row_count += 1 else: split_row_list.append(row_lines[:row_count]) break if row_count == len(row_lines): split_row_list.append(row_lines[:row_count]) break for col in col_lines: if col[3] <= y: col_count += 1 else: split_col_list.append(col_lines[:col_count]) break if col_count == len(col_lines): split_col_list.append(col_lines[:col_count]) break if row_count < len(row_lines) - 1: row_lines = row_lines[row_count:] split_row_list.append(row_lines) if col_count < len(col_lines) - 1: col_lines = col_lines[col_count:] split_col_list.append(col_lines) # print("len(split_row_list)", len(split_row_list)) # print("len(split_col_list)", len(split_col_list)) # 预测线取上下左右4个边(会有超出表格部分) [(), ()] area_row_line = [] area_col_line = [] for area in split_row_list: if not area: continue area.sort(key=lambda x: (x[1], x[0])) # print(area) up_line = area[0] bottom_line = area[-1] area_row_line.append([up_line, bottom_line]) for area in split_col_list: if not area: continue area.sort(key=lambda x: x[0]) left_line = area[0] right_line = area[-1] area_col_line.append([left_line, right_line]) # 线交点根据分割线纵坐标分为多个分割区域 points.sort(key=lambda x: (x[1], x[0])) point_count = 0 split_point_list = [] for y in new_split_y: points = points[point_count:len(points)] point_count = 0 for point in points: if point[1] <= y: point_count += 1 else: split_point_list.append(points[:point_count]) break if point_count == len(points): split_point_list.append(points[:point_count]) break if point_count < len(points) - 1: points = points[point_count:len(points)] split_point_list.append(points) # print("len(split_point_list)", len(split_point_list)) # 取每个分割区域的4条线(无超出表格部分) area_row_line2 = [] area_col_line2 = [] for area in split_point_list: if not area: continue area.sort(key=lambda x: (x[0], x[1])) left_up = area[0] right_bottom = area[-1] up_line = [left_up[0], left_up[1], right_bottom[0], left_up[1]] bottom_line = [left_up[0], right_bottom[1], right_bottom[0], right_bottom[1]] left_line = [left_up[0], left_up[1], left_up[0], right_bottom[1]] right_line = [right_bottom[0], left_up[1], right_bottom[0], right_bottom[1]] area_row_line2.append([up_line, bottom_line]) area_col_line2.append([left_line, right_line]) # 判断超出部分的长度，超出一定长度就补线 new_row_lines = [] new_col_lines = [] longer_row_lines = [] longer_col_lines = [] all_longer_row_lines = [] all_longer_col_lines = [] # print(len(area_col_line)) # print(len(area_col_line2)) for i in range(len(area_row_line)): up_line = area_row_line[i][0] up_line2 = area_row_line2[i][0] bottom_line = area_row_line[i][1] bottom_line2 = area_row_line2[i][1] left_line = area_col_line[i][0] left_line2 = area_col_line2[i][0] right_line = area_col_line[i][1] right_line2 = area_col_line2[i][1] # 补左右两条竖线超出来的线的row # print("left_line2[1] - left_line[1]", left_line2[1] - left_line[1]) if left_line2[1] - left_line[1] >= 10 and right_line2[1] - right_line[1] >= 10: if left_line2[1] - left_line[1] >= right_line2[1] - right_line[1]: new_row_lines.append([left_line[0], left_line[1], right_line[0], left_line[1]]) new_col_y = left_line[1] # 补了row，要将其他短的col连到row上 for col in split_col_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= col[1] <= col[3] <= bottom_line2[1]+3: longer_col_lines.append([col[0], min([new_col_y, col[1]]), col[2], col[3]]) else: new_row_lines.append([left_line[0], right_line[1], right_line[0], right_line[1]]) new_col_y = right_line[1] # 补了row，要将其他短的col连到row上 for col in split_col_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= col[1] <= col[3] <= bottom_line2[1]+3: longer_col_lines.append([col[0], min([new_col_y, col[1]]), col[2], col[3]]) # print("left_line[3] - left_line2[3]", left_line[3] - left_line2[3]) if left_line[3] - left_line2[3] >= 10 and right_line[3] - right_line2[3] >= 10: if left_line[3] - left_line2[3] >= right_line[3] - right_line2[3]: new_row_lines.append([left_line[2], left_line[3], right_line[2], left_line[3]]) new_col_y = left_line[3] # 补了row，要将其他短的col连到row上 for col in split_col_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= col[1] <= col[3] <= bottom_line2[1]+3: longer_col_lines.append([col[0], col[1], col[2], max([new_col_y, col[3]])]) else: new_row_lines.append([left_line[2], right_line[3], right_line[2], right_line[3]]) new_col_y = right_line[3] # 补了row，要将其他短的col连到row上 for col in split_col_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= col[1] <= col[3] <= bottom_line2[1]+3: longer_col_lines.append([col[0], col[1], col[2], max([new_col_y, col[3]])]) # 补上下两条横线超出来的线的col if up_line2[0] - up_line[0] >= 10 and bottom_line2[0] - bottom_line[0] >= 10: if up_line2[0] - up_line[0] >= bottom_line2[0] - bottom_line[0]: new_col_lines.append([up_line[0], up_line[1], up_line[0], bottom_line[1]]) new_row_x = up_line[0] # 补了col，要将其他短的row连到col上 for row in split_row_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= row[1] <= bottom_line2[1]+3: longer_row_lines.append([min([new_row_x, row[0]]), row[1], row[2], row[3]]) # new_row_lines.append([bottom_line[0], up_line[1], bottom_line[2], bottom_line[3]]) else: new_col_lines.append([bottom_line[0], up_line[1], bottom_line[0], bottom_line[1]]) new_row_x = bottom_line[0] # 补了col，要将其他短的row连到col上 for row in split_row_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= row[1] <= bottom_line2[1]+3: longer_row_lines.append([min([new_row_x, row[0]]), row[1], row[2], row[3]]) # new_row_lines.append([bottom_line[0], up_line[1], up_line[2], up_line[3]]) if up_line[2] - up_line2[2] >= 10 and bottom_line[2] - bottom_line2[2] >= 10: if up_line[2] - up_line2[2] >= bottom_line[2] - bottom_line2[2]: new_col_lines.append([up_line[2], up_line[3], up_line[2], bottom_line[3]]) new_row_x = up_line[2] # 补了col，要将其他短的row连到col上 for row in split_row_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= row[1] <= bottom_line2[1]+3: print([new_row_x, row[2]]) longer_row_lines.append([row[0], row[1], max([new_row_x, row[2]]), row[3]]) else: new_col_lines.append([bottom_line[2], up_line[3], bottom_line[2], bottom_line[3]]) new_row_x = bottom_line[2] # 补了col，要将其他短的row连到col上 for row in split_row_list[i]: # 需判断该线在这个区域中 if up_line2[1]-3 <= row[1] <= bottom_line2[1]+3: print([new_row_x, row[2]]) longer_row_lines.append([row[0], row[1], max([new_row_x, row[2]]), row[3]]) if longer_row_lines: all_longer_row_lines += longer_row_lines else: all_longer_row_lines += split_row_list[i] if longer_col_lines: all_longer_col_lines += longer_col_lines else: all_longer_col_lines += split_col_list[i] # 删除表格内部的补线 # temp_list = [] # for row in new_row_lines: # if up_line[1]-5 <= row[1] <= bottom_line[1]+5: # continue # temp_list.append(row) # print("fix_outline", new_row_lines) # new_row_lines = temp_list # print("fix_outline", new_row_lines) # temp_list = [] # for col in new_col_lines: # if left_line[0]-5 <= col[0] <= right_line[0]+5: # continue # temp_list.append(col) # # new_col_lines = temp_list # print("fix_outline", new_col_lines) # print("fix_outline", new_row_lines) # 删除重复包含的补线 # temp_list = [] # for row in new_row_lines: # if up_line[1]-5 <= row[1] <= bottom_line[1]+5: # continue # temp_list.append(row) # new_row_lines = temp_list # 展示上下左右边框线 # for i in range(len(area_row_line)): # print("row1", area_row_line[i]) # print("row2", area_row_line2[i]) # print("col1", area_col_line[i]) # print("col2", area_col_line2[i]) # cv2.line(image, (int(area_row_line[i][0][0]), int(area_row_line[i][0][1])), # (int(area_row_line[i][0][2]), int(area_row_line[i][0][3])), (0, 255, 0), 2) # cv2.line(image, (int(area_row_line2[i][1][0]), int(area_row_line2[i][1][1])), # (int(area_row_line2[i][1][2]), int(area_row_line2[i][1][3])), (0, 0, 255), 2) # cv2.imshow("fix_outline", image) # cv2.waitKey(0) return new_row_lines, new_col_lines, all_longer_row_lines, all_longer_col_lines def fix_table(row_point_list, col_point_list, split_y, row_lines, col_lines): # 分割线纵坐标 if len(split_y) < 2: return [] # 获取bbox bbox = [] # 每个点获取与其x最相近和y最相近的点 for i in range(1, len(split_y)): # 循环每行 for row in row_point_list: row.sort(key=lambda x: (x[0], x[1])) # 行不在该区域跳过 if row[0][1] <= split_y[i-1] or row[0][1] >= split_y[i]: continue # print("len(row)", len(row)) # print("row", row) # 循环行中的点 for j in range(len(row)): if j == len(row) - 1: break current_point = row[j] next_point_in_row_list = row[j+1:] # 循环这一行的下一个点 for next_point_in_row in next_point_in_row_list: # 是否在这一行点找到，找不到就这一行的下个点 not_found = 1 # 查询下个点所在列 next_col = [] for col in col_point_list: col.sort(key=lambda x: (x[1], x[0])) # 列不在该区域跳过 if col[0][1] <= split_y[i-1] or col[-1][1] >= split_y[i]: continue if col[0][0]-3 <= next_point_in_row[0] <= col[0][0]+3: next_col = col break # 循环匹配当前点和下一列点 next_col.sort(key=lambda x: (x[1], x[0])) for point1 in next_col: # 同一行的就跳过 if current_point[1]-3 <= point1[1] <= current_point[1]+3: continue if point1[1] <= current_point[1]-3: continue # 候选bbox candidate_bbox = [current_point[0], current_point[1], point1[0], point1[1]] # print("candidate_bbox", candidate_bbox) # 判断该bbox是否存在，判断bbox的上下两条边是否有包含在row中 contain_flag1 = 0 contain_flag2 = 0 for row1 in row_lines: # 行不在该区域跳过 if row1[1] <= split_y[i-1] or row1[1] >= split_y[i]: continue # bbox上边框 y一样 if not contain_flag1: if row1[1]-3 <= candidate_bbox[1] <= row1[1]+3: # 格子里的断开线段 row1_break = (max([row1[0], candidate_bbox[0]]), row1[1], min([row1[2], candidate_bbox[2]]), row1[3]) if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: contain_flag1 = 1 # bbox下边框 y一样 if not contain_flag2: if row1[1]-3 <= candidate_bbox[3] <= row1[1]+3: # 格子里的断开线段 row1_break = (max([row1[0], candidate_bbox[0]]), row1[1], min([row1[2], candidate_bbox[2]]), row1[3]) if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: contain_flag2 = 1 # 判断该bbox是否存在，判断bbox的左右两条边是否有包含在col中 contain_flag3 = 0 contain_flag4 = 0 for col1 in col_lines: # 列不在该区域跳过 if col1[1] <= split_y[i-1] or col1[3] >= split_y[i]: continue # bbox左边线 x一样 if not contain_flag3: if col1[0]-3 <= candidate_bbox[0] <= col1[0]+3: # 格子里的断开线段 col1_break = (col1[0], max([col1[1], candidate_bbox[1]]), col1[2], min([col1[3], candidate_bbox[3]])) if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: contain_flag3 = 1 # bbox右边框 x一样 if not contain_flag4: if col1[0]-3 <= candidate_bbox[2] <= col1[0]+3: # 格子里的断开线段 col1_break = (col1[0], max([col1[1], candidate_bbox[1]]), col1[2], min([col1[3], candidate_bbox[3]])) if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: contain_flag4 = 1 # 找到了该bbox，并且是存在的 if contain_flag1 and contain_flag2 and contain_flag3 and contain_flag4: bbox.append([(candidate_bbox[0], candidate_bbox[1]), (candidate_bbox[2], candidate_bbox[3])]) not_found = 0 break if not not_found: break return bbox def delete_close_points(point_list, row_point_list, col_point_list, threshold=5): new_point_list = [] delete_point_list = [] point_list.sort(key=lambda x: (x[1], x[0])) for i in range(len(point_list)): point1 = point_list[i] if point1 in delete_point_list: continue if i == len(point_list) - 1: new_point_list.append(point1) break point2 = point_list[i+1] # 判断坐标 if abs(point1[0] - point2[0]) > threshold or abs(point1[1] - point2[1]) > threshold: new_point_list.append(point1) else: # 看两个点上的相同坐标点哪个多，就保留哪个 count1 = 0 count2 = 0 for col in col_point_list: if point1[0] == col[0][0]: count1 += len(col) elif point2[0] == col[0][0]: count2 += len(col) if count1 >= count2: new_point_list.append(point1) delete_point_list.append(point2) else: new_point_list.append(point2) delete_point_list.append(point1) point_list = new_point_list new_point_list = [] delete_point_list = [] point_list.sort(key=lambda x: (x[0], x[1])) for i in range(len(point_list)): point1 = point_list[i] if point1 in delete_point_list: continue if i == len(point_list) - 1: new_point_list.append(point1) break point2 = point_list[i+1] # 判断坐标 if abs(point1[0] - point2[0]) > threshold or abs(point1[1] - point2[1]) > threshold: new_point_list.append(point1) else: count1 = 0 count2 = 0 for row in row_point_list: if point1[0] == row[0][0]: count1 += len(row) elif point2[0] == row[0][0]: count2 += len(row) if count1 >= count2: new_point_list.append(point1) delete_point_list.append(point2) else: new_point_list.append(point2) delete_point_list.append(point1) return new_point_list def get_bbox2(image_np, points): # # 坐标点按行分 # row_point_list = [] # row_point = [] # points.sort(key=lambda x: (x[0], x[1])) # for p in points: # if len(row_point) == 0: # x = p[0] # if x-5 <= p[0] <= x+5: # row_point.append(p) # else: # row_point_list.append(row_point) # row_point = [] # # 坐标点按列分 # col_point_list = [] # col_point = [] # points.sort(key=lambda x: (x[1], x[0])) # for p in points: # if len(col_point) == 0: # y = p[1] # if y-5 <= p[1] <= y+5: # col_point.append(p) # else: # col_point_list.append(col_point) # col_point = [] row_point_list = get_points_row(points) col_point_list = get_points_col(points) print("len(points)", len(points)) for point in points: cv2.circle(image_np, point, 1, (0, 255, 0), 1) cv2.imshow("points_deleted", image_np) points = delete_close_points(points, row_point_list, col_point_list) print("len(points)", len(points)) for point in points: cv2.circle(image_np, point, 1, (255, 0, 0), 3) cv2.imshow("points_deleted", image_np) cv2.waitKey(0) row_point_list = get_points_row(points, 5) col_point_list = get_points_col(points, 5) print("len(row_point_list)", len(row_point_list)) for row in row_point_list: print("row", len(row)) print("col_point_list", len(col_point_list)) for col in col_point_list: print("col", len(col)) bbox = [] for i in range(len(row_point_list)): if i == len(row_point_list) - 1: break # 遍历每个row的point，找到其所在列的下一个点和所在行的下一个点 current_row = row_point_list[i] for j in range(len(current_row)): current_point = current_row[j] if j == len(current_row) - 1: break next_row_point = current_row[j+1] # 找出当前点所在的col，得到该列下一个point current_col = col_point_list[j] for k in range(len(current_col)): if current_col[k][1] > current_point[1] + 10: next_col_point = current_col[k] break next_row = row_point_list[k] for k in range(len(next_row)): if next_row[k][0] >= next_row_point[0] + 5: next_point = next_row[k] break # 得到bbox bbox.append([(current_point[0], current_point[1]), (next_point[0], next_point[1])]) # bbox = [] # for p in points: # # print("p", p) # p_row = [] # p_col = [] # for row in row_point_list: # if p[0] == row[0][0]: # for p1 in row: # if abs(p[1]-p1[1]) <= 5: # continue # p_row.append([p1, abs(p[1]-p1[1])]) # p_row.sort(key=lambda x: x[1]) # for col in col_point_list: # if p[1] == col[0][1]: # for p2 in col: # if abs(p[0]-p2[0]) <= 5: # continue # p_col.append([p2, abs(p[0]-p2[0])]) # p_col.sort(key=lambda x: x[1]) # if len(p_row) == 0 or len(p_col) == 0: # continue # break_flag = 0 # for i in range(len(p_row)): # for j in range(len(p_col)): # # print(p_row[i][0]) # # print(p_col[j][0]) # another_point = (p_col[j][0][0], p_row[i][0][1]) # # print("another_point", another_point) # if abs(p[0]-another_point[0]) <= 5 or abs(p[1]-another_point[1]) <= 5: # continue # if p[0] >= another_point[0] or p[1] >= another_point[1]: # continue # if another_point in points: # box = [p, another_point] # box.sort(key=lambda x: x[0]) # if box not in bbox: # bbox.append(box) # break_flag = 1 # break # if break_flag: # break # # # delete duplicate # delete_bbox = [] # for i in range(len(bbox)): # for j in range(i+1, len(bbox)): # if bbox[i][0] == bbox[j][0]: # if bbox[i][1][0] - bbox[j][1][0] <= 3 \ # and bbox[i][1][1] - bbox[j][1][1] <= 3: # delete_bbox.append(bbox[j]) # if bbox[i][1] == bbox[j][1]: # if bbox[i][0][0] - bbox[j][0][0] <= 3 \ # and bbox[i][0][1] - bbox[j][0][1] <= 3: # delete_bbox.append(bbox[j]) # # delete too small area # # for box in bbox: # # if box[1][0] - box[0][0] <= # for d_box in delete_bbox: # if d_box in bbox: # bbox.remove(d_box) # print bbox bbox.sort(key=lambda x: (x[0][0], x[0][1], x[1][0], x[1][1])) # origin bbox # origin_bbox = [] # for box in bbox: # origin_bbox.append([(box[0][0], box[0][1] - 40), (box[1][0], box[1][1] - 40)]) # for box in origin_bbox: # cv2.rectangle(origin_image, box[0], box[1], (0, 0, 255), 2, 8) # cv2.imshow('AlanWang', origin_image) # cv2.waitKey(0) for box in bbox: cv2.rectangle(image_np, box[0], box[1], (0, 0, 255), 2, 8) cv2.imshow('bboxes', image_np) cv2.waitKey(0) # for point in points: # print(point) # cv2.circle(image_np, point, 1, (0, 0, 255), 3) # cv2.imshow('points', image_np) # cv2.waitKey(0) return bbox def get_bbox1(image_np, points, split_y): # 分割线纵坐标 # print("split_y", split_y) if len(split_y) < 2: return [] # 计算行列，剔除相近交点 row_point_list = get_points_row(points) col_point_list = get_points_col(points) print("len(row_point_list)", row_point_list) print("len(col_point_list)", len(col_point_list)) # for point in points: # cv2.circle(image_np, point, 1, (0, 255, 0), 1) # cv2.imshow("points", image_np) points = delete_close_points(points, row_point_list, col_point_list) # print("len(points)", len(points)) # for point in points: # cv2.circle(image_np, point, 1, (255, 0, 0), 3) # cv2.imshow("points_deleted", image_np) # cv2.waitKey(0) # 获取bbox bbox = [] # 每个点获取与其x最相近和y最相近的点 for i in range(1, len(split_y)): for point1 in points: if point1[1] <= split_y[i-1] or point1[1] >= split_y[i]: continue distance_x = 10000 distance_y = 10000 x = 0 y = 0 threshold = 10 for point2 in points: if point2[1] <= split_y[i-1] or point2[1] >= split_y[i]: continue # 最近 x y if 2 < point2[0] - point1[0] < distance_x and point2[1] - point1[1] <= threshold: distance_x = point2[0] - point1[0] x = point2[0] if 2 < point2[1] - point1[1] < distance_y and point2[0] - point1[0] <= threshold: distance_y = point2[1] - point1[1] y = point2[1] if not x or not y: continue bbox.append([(point1[0], point1[1]), (x, y)]) # 删除包含关系bbox temp_list = [] for i in range(len(bbox)): box1 = bbox[i] for j in range(len(bbox)): if i == j: continue box2 = bbox[j] contain_flag = 0 if box2[0][0] <= box1[0][0] <= box1[1][0] <= box2[1][0] and \ box2[0][1] <= box1[0][1] <= box1[1][1] <= box2[1][1]: contain_flag = 1 break temp_list.append(box1) bbox = temp_list # 展示 for box in bbox: # print(box[0], box[1]) # if abs(box[0][1] - box[1][1]) > abs(box[0][0] - box[1][0]): # continue cv2.rectangle(image_np, box[0], box[1], (0, 0, 255), 2, 8) cv2.imshow('bboxes', image_np) cv2.waitKey(0) return bbox def get_bbox0(image_np, row_point_list, col_point_list, split_y, row_lines, col_lines): # 分割线纵坐标 if len(split_y) < 2: return [] # 计算行列，剔除相近交点 # row_point_list = get_points_row(points) # col_point_list = get_points_col(points) # points = delete_close_points(points, row_point_list, col_point_list) # row_point_list = get_points_row(points) # col_point_list = get_points_col(points) # 获取bbox bbox = [] # print("get_bbox split_y", split_y) # 每个点获取与其x最相近和y最相近的点 for i in range(1, len(split_y)): # 循环每行 for row in row_point_list: row.sort(key=lambda x: (x[0], x[1])) # 行不在该区域跳过 if row[0][1] <= split_y[i-1] or row[0][1] >= split_y[i]: continue # 循环行中的点 for j in range(len(row)): if j == len(row) - 1: break current_point = row[j] next_point_in_row = row[j+1] # 查询下个点所在列 next_col = [] for col in col_point_list: col.sort(key=lambda x: (x[1], x[0])) # 列不在该区域跳过 if col[0][1] <= split_y[i-1] or col[-1][1] >= split_y[i]: continue if col[0][0]-3 <= next_point_in_row[0] <= col[0][0]+3: next_col = col break # 循环匹配当前点和下一列点 for point1 in next_col: # 同一行的就跳过 if current_point[1]-3 <= point1[1] <= current_point[1]+3: continue if point1[1] <= current_point[1]-3: continue # 候选bbox candidate_bbox = [current_point[0], current_point[1], point1[0], point1[1]] # 判断该bbox是否存在，线条包含关系 contain_flag1 = 0 contain_flag2 = 0 for row1 in row_lines: # 行不在该区域跳过 if row1[1] <= split_y[i-1] or row1[1] >= split_y[i]: continue # bbox上边框 y一样 if not contain_flag1: if row1[1]-3 <= candidate_bbox[1] <= row1[1]+3: # candidate的x1,x2需被包含在row线中 if row1[0]-3 <= candidate_bbox[0] <= candidate_bbox[2] <= row1[2]+3: contain_flag1 = 1 # bbox下边框 y一样 if not contain_flag2: if row1[1]-3 <= candidate_bbox[3] <= row1[1]+3: # candidate的x1,x2需被包含在row线中 if row1[0]-3 <= candidate_bbox[0] <= candidate_bbox[2] <= row1[2]+3: contain_flag2 = 1 # 找到了该bbox，并且是存在的 if contain_flag1 and contain_flag2: bbox.append([(candidate_bbox[0], candidate_bbox[1]), (candidate_bbox[2], candidate_bbox[3])]) break return bbox def get_bbox(image_np, row_point_list, col_point_list, split_y, row_lines, col_lines): # 分割线纵坐标 if len(split_y) < 2: return [] # 获取bbox bbox = [] # 每个点获取与其x最相近和y最相近的点 for i in range(1, len(split_y)): # 循环每行 for row in row_point_list: row.sort(key=lambda x: (x[0], x[1])) # 行不在该区域跳过 if row[0][1] <= split_y[i-1] or row[0][1] >= split_y[i]: continue # print("len(row)", len(row)) # print("row", row) # 循环行中的点 for j in range(len(row)): if j == len(row) - 1: break current_point = row[j] next_point_in_row_list = row[j+1:] # 循环这一行的下一个点 for next_point_in_row in next_point_in_row_list: # 是否在这一行点找到，找不到就这一行的下个点 not_found = 1 # 查询下个点所在列 next_col = [] for col in col_point_list: col.sort(key=lambda x: (x[1], x[0])) # 列不在该区域跳过 if col[0][1] <= split_y[i-1] or col[-1][1] >= split_y[i]: continue if col[0][0]-3 <= next_point_in_row[0] <= col[0][0]+3: next_col = col break # 循环匹配当前点和下一列点 next_col.sort(key=lambda x: (x[1], x[0])) for point1 in next_col: # 同一行的就跳过 if current_point[1]-3 <= point1[1] <= current_point[1]+3: continue if point1[1] <= current_point[1]-3: continue # 候选bbox candidate_bbox = [current_point[0], current_point[1], point1[0], point1[1]] # print("candidate_bbox", candidate_bbox) # 判断该bbox是否存在，判断bbox的上下两条边是否有包含在row中 contain_flag1 = 0 contain_flag2 = 0 for row1 in row_lines: # 行不在该区域跳过 if row1[1] <= split_y[i-1] or row1[1] >= split_y[i]: continue # bbox上边框 y一样 if not contain_flag1: if row1[1]-3 <= candidate_bbox[1] <= row1[1]+3: # 格子里的断开线段 row1_break = (max([row1[0], candidate_bbox[0]]), row1[1], min([row1[2], candidate_bbox[2]]), row1[3]) if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: contain_flag1 = 1 # # candidate的x1,x2需被包含在row线中 # if row1[0]-3 <= candidate_bbox[0] <= candidate_bbox[2] <= row1[2]+3: # contain_flag1 = 1 # # # 判断线条有无端点在格子中 # elif candidate_bbox[0] < row1[0] < candidate_bbox[2] \ # or candidate_bbox[0] < row1[2] < candidate_bbox[2]: # # 线条会有缺一点情况，判断长度超过格子一半 # if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: # contain_flag1 = 1 # bbox下边框 y一样 if not contain_flag2: if row1[1]-3 <= candidate_bbox[3] <= row1[1]+3: # 格子里的断开线段 row1_break = (max([row1[0], candidate_bbox[0]]), row1[1], min([row1[2], candidate_bbox[2]]), row1[3]) if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: contain_flag2 = 1 # # candidate的x1,x2需被包含在row线中 # if row1[0]-3 <= candidate_bbox[0] <= candidate_bbox[2] <= row1[2]+3: # contain_flag2 = 1 # # # 判断线条有无端点在格子中 # elif candidate_bbox[0] < row1[0] < candidate_bbox[2] \ # or candidate_bbox[0] < row1[2] < candidate_bbox[2]: # # 线条会有缺一点情况，判断长度超过格子一半 # if row1_break[2] - row1_break[0] >= (candidate_bbox[2] - candidate_bbox[0])/3: # contain_flag2 = 1 # 判断该bbox是否存在，判断bbox的左右两条边是否有包含在col中 contain_flag3 = 0 contain_flag4 = 0 for col1 in col_lines: # 列不在该区域跳过 if col1[1] <= split_y[i-1] or col1[3] >= split_y[i]: continue # bbox左边线 x一样 if not contain_flag3: if col1[0]-3 <= candidate_bbox[0] <= col1[0]+3: # 格子里的断开线段 col1_break = (col1[0], max([col1[1], candidate_bbox[1]]), col1[2], min([col1[3], candidate_bbox[3]])) if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: contain_flag3 = 1 # # candidate的y1,y2需被包含在col线中 # if col1[1]-3 <= candidate_bbox[1] <= candidate_bbox[3] <= col1[3]+3: # contain_flag3 = 1 # # # 判断线条有无端点在格子中 # elif candidate_bbox[1] < col1[1] < candidate_bbox[3] \ # or candidate_bbox[1] < col1[3] < candidate_bbox[3]: # # 线条会有缺一点情况，判断长度超过格子一半 # if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: # contain_flag3 = 1 # bbox右边框 x一样 if not contain_flag4: if col1[0]-3 <= candidate_bbox[2] <= col1[0]+3: # 格子里的断开线段 col1_break = (col1[0], max([col1[1], candidate_bbox[1]]), col1[2], min([col1[3], candidate_bbox[3]])) if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: contain_flag4 = 1 # # candidate的y1,y2需被包含在col线中 # if col1[1]-3 <= candidate_bbox[1] <= candidate_bbox[3] <= col1[3]+3: # contain_flag4 = 1 # # # 判断线条有无端点在格子中 # elif candidate_bbox[1] < col1[1] < candidate_bbox[3] \ # or candidate_bbox[1] < col1[3] < candidate_bbox[3]: # # 线条会有缺一点情况，判断长度超过格子一半 # if col1_break[3] - col1_break[1] >= (candidate_bbox[3] - candidate_bbox[1])/3: # contain_flag4 = 1 # 找到了该bbox，并且是存在的 if contain_flag1 and contain_flag2 and contain_flag3 and contain_flag4: bbox.append([(candidate_bbox[0], candidate_bbox[1]), (candidate_bbox[2], candidate_bbox[3])]) not_found = 0 # print("candidate_bbox", candidate_bbox) # print(contain_flag1, contain_flag2, contain_flag3, contain_flag4) break # else: # print("candidate_bbox", candidate_bbox) # print(contain_flag1, contain_flag2, contain_flag3, contain_flag4) if not not_found: break return bbox def get_bbox_by_contours(image_np): img_gray = cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY) ret, img_bin = cv2.threshold(img_gray, 127, 255, cv2.THRESH_BINARY) # 3.连通域分析 img_bin, contours, hierarchy = cv2.findContours(img_bin, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE) # 4.获取最小外接圆 圆心 半径 center, radius = cv2.minEnclosingTriangle(contours[0]) center = np.int0(center) # 5.绘制最小外接圆 img_result = image_np.copy() cv2.circle(img_result, tuple(center), int(radius), (255, 255, 255), 2) # # 读入图片 # img = image_np # cv2.imshow("get_bbox_by_contours ", image_np) # # 中值滤波，去噪 # img = cv2.medianBlur(img, 3) # gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # cv2.namedWindow('original', cv2.WINDOW_AUTOSIZE) # cv2.imshow('original', gray) # # # 阈值分割得到二值化图片 # ret, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY | cv2.THRESH_OTSU) # # # 膨胀操作 # kernel2 = cv2.getStructuringElement(cv2.MORPH_RECT, (3, 3)) # bin_clo = cv2.dilate(binary, kernel2, iterations=2) # # # 连通域分析 # num_labels, labels, stats, centroids = cv2.connectedComponentsWithStats(bin_clo, connectivity=8) # # # 查看各个返回值 # # 连通域数量 # print('num_labels = ',num_labels) # # 连通域的信息：对应各个轮廓的x、y、width、height和面积 # print('stats = ',stats) # # 连通域的中心点 # print('centroids = ',centroids) # # 每一个像素的标签1、2、3.。。，同一个连通域的标签是一致的 # print('labels = ',labels) # # # 不同的连通域赋予不同的颜色 # output = np.zeros((img.shape[0], img.shape[1], 3), np.uint8) # for i in range(1, num_labels): # # mask = labels == i # output[:, :, 0][mask] = np.random.randint(0, 255) # output[:, :, 1][mask] = np.random.randint(0, 255) # output[:, :, 2][mask] = np.random.randint(0, 255) # cv2.imshow('oginal', output) # cv2.waitKey() # cv2.destroyAllWindows() def get_points_col(points, split_y, threshold=5): # 坐标点按行分 row_point_list = [] row_point = [] points.sort(key=lambda x: (x[0], x[1])) x = points[0][0] for i in range(1, len(split_y)): for p in points: if p[1] <= split_y[i-1] or p[1] >= split_y[i]: continue if x-threshold <= p[0] <= x+threshold: row_point.append(p) else: row_point.sort(key=lambda x: (x[1], x[0])) if row_point: row_point_list.append(row_point) row_point = [] x = p[0] row_point.append(p) if row_point: row_point_list.append(row_point) return row_point_list def get_points_row(points, split_y, threshold=5): # 坐标点按列分 col_point_list = [] col_point = [] points.sort(key=lambda x: (x[1], x[0])) y = points[0][1] for i in range(len(split_y)): for p in points: if p[1] <= split_y[i-1] or p[1] >= split_y[i]: continue if y-threshold <= p[1] <= y+threshold: col_point.append(p) else: col_point.sort(key=lambda x: (x[0], x[1])) if col_point: col_point_list.append(col_point) col_point = [] y = p[1] col_point.append(p) if col_point: col_point_list.append(col_point) return col_point_list def get_outline_point(points, split_y): # 分割线纵坐标 # print("get_outline_point split_y", split_y) if len(split_y) < 2: return [] outline_2point = [] points.sort(key=lambda x: (x[1], x[0])) for i in range(1, len(split_y)): area_points = [] for point in points: if point[1] <= split_y[i-1] or point[1] >= split_y[i]: continue area_points.append(point) if area_points: area_points.sort(key=lambda x: (x[1], x[0])) outline_2point.append([area_points[0], area_points[-1]]) return outline_2point # def merge_row(row_lines): # for row in row_lines: # for row1 in row_lines: def get_best_predict_size(image_np): sizes = [1280, 1152, 1024, 896, 768, 640, 512, 384, 256, 128] min_len = 10000 best_height = sizes[0] for height in sizes: if abs(image_np.shape[0] - height) < min_len: min_len = abs(image_np.shape[0] - height) best_height = height min_len = 10000 best_width = sizes[0] for width in sizes: if abs(image_np.shape[1] - width) < min_len: min_len = abs(image_np.shape[1] - width) best_width = width return best_height, best_width def choose_longer_row(lines): new_row = [] jump_row = [] for i in range(len(lines)): row1 = lines[i] jump_flag = 0 if row1 in jump_row: continue for j in range(i+1, len(lines)): row2 = lines[j] if row2 in jump_row: continue if row2[1]-5 <= row1[1] <= row2[1]+5: if row1[0] <= row2[0] and row1[2] >= row2[2]: new_row.append(row1) jump_row.append(row1) jump_row.append(row2) jump_flag = 1 break elif row2[0] <= row1[0] and row2[2] >= row1[2]: new_row.append(row2) jump_row.append(row1) jump_row.append(row2) jump_flag = 1 break if not jump_flag: new_row.append(row1) jump_row.append(row1) return new_row def choose_longer_col(lines): new_col = [] jump_col = [] for i in range(len(lines)): col1 = lines[i] jump_flag = 0 if col1 in jump_col: continue for j in range(i+1, len(lines)): col2 = lines[j] if col2 in jump_col: continue if col2[0]-5 <= col1[0] <= col2[0]+5: if col1[1] <= col2[1] and col1[3] >= col2[3]: new_col.append(col1) jump_col.append(col1) jump_col.append(col2) jump_flag = 1 break elif col2[1] <= col1[1] and col2[3] >= col1[3]: new_col.append(col2) jump_col.append(col1) jump_col.append(col2) jump_flag = 1 break if not jump_flag: new_col.append(col1) jump_col.append(col1) return new_col def line_fix(image_np): image_binary = cv2.cvtColor(image_np, cv2.COLOR_BGR2GRAY) rows, cols = image_binary.shape scale = 100 kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (1, rows//scale)) row_erode = cv2.erode(image_binary, kernel, iterations=1) cv2.imshow("row_erode", row_erode) scale = 100 kernel = cv2.getStructuringElement(cv2.MORPH_RECT, (cols//scale, 1)) col_erode = cv2.erode(image_binary, kernel, iterations=1) cv2.imshow("col_erode", col_erode) image_erode = row_erode + col_erode cv2.imshow("image_erode", image_erode) return image_erode if __name__ == '__main__': net = table_net((1024, 768, 3), 2) net.summary()