如何提升pytesseract识别在线图片数字的准确率？

我需要从以下股票图表图片中提取数字数据，但pytesseract识别效果极差，希望得到有效优化方案：

已尝试的代码及问题 #

初始代码 #

import io
import requests
import pytesseract
from PIL import Image
response = requests.get("https://port.jpx.co.jp/jpx/chart/chart21.exe?template=ini/DayIndexCSV&basequote=151_2024&begin=2024/4/2&end=2024/04/02&mode=D")
img = Image.open(io.BytesIO(response.content))
text = pytesseract.image_to_string(img)
print(text)
img.show()

#Updated black and white version
img.save('1.png')
image = cv2.imread('1.png',0)
_, thresh1 = cv2.threshold(image, 105, 255, cv2.THRESH_BINARY)
_, thresh2 = cv2.threshold(image, 106, 255, cv2.THRESH_BINARY_INV)
final_thresh = cv2.bitwise_and(thresh1, thresh2)
im = Image.fromarray(final_thresh)
text = pytesseract.image_to_string(im)
print(text)
im.show()

识别问题：大量数字（如2、4、5、6）被识别为“8”，分隔符有时识别为“.”有时为“,”。

裁剪聚焦区域尝试 #

w, h= img.size
img2=img.crop((240, 185, w-220, h-220))
text = pytesseract.image_to_string(img2)
print(text)

该区域真实值为“2,714.45”，但返回“eT AS”；添加配置后：

text = pytesseract.image_to_string(img2, config='--psm 10 --oem 3 -c tessedit_char_whitelist=0123456789')
print(text)

仅返回“1”。

调整DPI与阈值后的代码 #

from PIL import Image
import pytesseract
import requests
import io

response = requests.get("https://port.jpx.co.jp/jpx/chart/chart21.exe?template=ini/DayIndexCSV&basequote=151_2024&begin=2024/4/02&end=2024/04/02&mode=D")
img = Image.open(io.BytesIO(response.content))
grayscale_image = img.convert("L")
original_width, original_height = grayscale_image.size
factor = 300/72 # enlarge from default 72 to 300 dpi
new_width = int(original_width * factor)
new_height = int(original_height * factor)
enlarged_image = grayscale_image.resize((new_width, new_height))
threshold = 175 # found by trial/error (removes disturbing grid lines)
bw_image = enlarged_image.point(lambda x: 0 if x < threshold else 255, '1') # mode='1' -> b/w 

w, h= bw_image.size
bw_image=bw_image.crop((500, 0, w-300, h-480))
text = pytesseract.image_to_string(bw_image)
print(text)
bw_image.show()

输出结果仍有错误：

2U2d Ud Ue
2714.45

271445
2024/04/02

对应的处理后图片：

希望得到能精准识别图片中数字及日期的优化方案。

内容的提问来源于stack exchange，提问作者Wick