图片识别文字之图片中文字的分割思路

一张图片中一大堆字符，如何识别？思路如下：1，先分割成单个字符；2，再将每个字符独立识别。3，组合已识别的单个文字；4，“语义”修正最后的识别结果。

Python图片分割图片的示例（原文：https://www.jianshu.com/p/64808391285e）

其思路：使用cv2读取图片文件，然后二值化，根据垂直和水平方向像素与空间阈值确定方格分割文字，再分别读取每个已经分割的文字。

代码如下：

import os
import cv2
import numpy as np

base_dir = "./origin/"
dst_dir = "./result/"
min_val = 10
min_range = 30

count = 0
def extract_peek(array_vals, minimun_val, minimun_range):
    start_i = None
    end_i = None
    peek_ranges = []
    for i, val in enumerate(array_vals):
        if val > minimun_val and start_i is None:
            start_i = i
        elif val > minimun_val and start_i is not None:
            pass
        elif val < minimun_val and start_i is not None:
            if i - start_i >= minimun_range:
                end_i = i
                print(end_i - start_i)
                peek_ranges.append((start_i, end_i))
                start_i = None
                end_i = None
        elif val < minimun_val and start_i is None:
            pass
        else:
            raise ValueError("cannot parse this case...")
    return peek_ranges
def cutImage(img, peek_range):
    global count
    for i, peek_range in enumerate(peek_ranges):
        for vertical_range in vertical_peek_ranges2d[i]:
            x = vertical_range[0]
            y = peek_range[0]
            w = vertical_range[1] - x
            h = peek_range[1] - y
            pt1 = (x, y)
            pt2 = (x + w, y + h)
            count += 1
            img1 = img[y:peek_range[1], x:vertical_range[1]]
            new_shape = (150, 150)
            img1 = cv2.resize(img1, new_shape)
            cv2.imwrite(dst_dir + str(count) + ".png", img1)
            # cv2.rectangle(img, pt1, pt2, color)
for fileName in os.listdir(base_dir):
    img = cv2.imread(base_dir + fileName)
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    adaptive_threshold = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_GAUSSIAN_C, \
                                               cv2.THRESH_BINARY_INV, 11, 2)
    horizontal_sum = np.sum(adaptive_threshold, axis=1)
    peek_ranges = extract_peek(horizontal_sum, min_val, min_range)
    line_seg_adaptive_threshold = np.copy(adaptive_threshold)
    for i, peek_range in enumerate(peek_ranges):
        x = 0
        y = peek_range[0]
        w = line_seg_adaptive_threshold.shape[1]
        h = peek_range[1] - y
        pt1 = (x, y)
        pt2 = (x + w, y + h)
        cv2.rectangle(line_seg_adaptive_threshold, pt1, pt2, 255)
    vertical_peek_ranges2d = []
    for peek_range in peek_ranges:
        start_y = peek_range[0]
        end_y = peek_range[1]
        line_img = adaptive_threshold[start_y:end_y, :]
        vertical_sum = np.sum(line_img, axis=0)
        vertical_peek_ranges = extract_peek(
            vertical_sum, min_val, min_range)
        vertical_peek_ranges2d.append(vertical_peek_ranges)
    cutImage(img, peek_range)