new_tiku_structure_v3_sci/utils/img2latex.py

import base64
import json
import random
import requests
import configs
import cv2
import numpy as np


def np2base64(img_arr):
    """

    :param img_arr:
    :return:
    """
    retval, buffer = cv2.imencode('.jpg', img_arr)
    base64_str = base64.b64encode(buffer)
    base64_str = base64_str.decode()
    return base64_str

def microsoft_ocr_one(img_base64):
    """
    :param img_base64:
    :return:
    """
    data = {"data": img_base64,
            "inputForm": "Image",
            "clientInfo": {"app": "Math",
                           "platform": "ios",
                           "configuration": "Unknown",
                           "version": "1.8.0",
                           "mkt": "zh-cn"},
            }
    url = "https://www.bing.com/cameraexp/api/v1/getlatex"
    headers = {"User-Agent": "Math/1 CFNetwork/1121.2.2 Darwin/19.3.0",
               # $(CFBundleName)/$(CFBundleVersion) CFNetwork/1121.2.2 Darwin/19.2.0
               "Content-Type": "application/json",
               "Connection": "keep-alive",
               "Accept": "application/json",
               "Content-Length": "888888",
               "Accept-Language": "zh-cn",
               "Accept-Encoding": "gzip, deflate, br"
               }
    resp = requests.post(url=url, headers=headers, data=json.dumps(data, ensure_ascii=True), timeout=3)
    ocrres = ""
    if resp.status_code == 200:
        ocrres = json.loads(resp.text)["latex"]
        print(resp.text)
    return ocrres


def formula_reg(img_base64):
    """

    :param img_base64:
    :return: result 包括参数：
                "image_url": img_path，输入图片的名称 ;
                "texts": 识别结果;
                "is_success": 是否识别成功  --- 0：识别错误  1：识别正确 99：接口调用错误
    """
    flag = random.randint(1, 100)
    if flag <= 90:
        # mathpix charged api
        print("---mathpix---")
        txt_res = requests.post(url=configs.mathpix_ip,
                                data={"img_url": img_base64}, timeout=3).json()["texts"]
        txt_res = txt_res.replace("<latex>", "").replace("</latex>", "")
    else:  # 11-15 p=1/3 free api
        txt_res = microsoft_ocr_one(img_base64)
    return txt_res


def get_ocr_latex(img_path):
    """
    根据公式图片调取接口（mathpix,微软），获取latex
    img_path:图片路径
    :return:
    """
    try:
        #
        img = cv2.imread(img_path, cv2.IMREAD_UNCHANGED)
        rr = cv2.split(img)
        # print(img.shape)
        # img = img[:, :, :3]
        base64_img = np2base64(rr[3])
        res_ltx = formula_reg(base64_img)
    except:
        res_ltx = ""
    return res_ltx


def get_ocrlatex_by_url(img_url):
    """
    根据公式图片调取接口（mathpix,微软），获取latex
    img_path:图片路径
    :return:
    """
    try:
        img = requests.get(img_url).content
        im = np.frombuffer(img, np.uint8)
        img = cv2.imdecode(im, cv2.IMREAD_UNCHANGED)
        rr = cv2.split(img)
        # print(img.shape)
        # img = img[:, :, :3]
        base64_img = np2base64(rr[3])
        res_ltx = formula_reg(base64_img)
        print(res_ltx)
    except:
        res_ltx = ""
    return res_ltx


if __name__ == "__main__":
    import time
    # t1 = time.time()
    # hyy = "http://192.168.1.140:8888/ser_static/1848/files/image337.png"
    # res = get_ocrlatex_by_url(hyy)
    # print(res)
    # print(time.time()-t1)
    # F:\zwj\word_folder\6673\word\media\image14.png  # wmf生成的4通道
    # F:\zwj\word_folder\6210\word\media\image47.png  # 原始4通道图片

    # im = requests.get(hyy).content
    # img1 = np.frombuffer(im, np.uint8)
    # img = cv2.imread(r"F:\zwj\word_folder\6673\word\media\image14.png", cv2.IMREAD_UNCHANGED)
    # img = cv2.imdecode(img1, cv2.IMREAD_UNCHANGED)
    # rr = cv2.split(img)
    # print(img.shape)
    # img = img[:, :, :3]
    # import numpy as np
    # # print(np.argwhere((0 < img) & (img <255)))
    # cv2.imshow("hh", img)
    # cv2.waitKey(0)
    # self.raw = cv2.imdecode(data, cv2.IMREAD_COLOR)
    # img = cv2.resize(rr[3], (0, 0), fx=0.5, fy=0.3)
    # cv2.imshow("hh", img)
    # cv2.waitKey(0)
    # base64_img = np2base64(rr[3])
    # # res = microsoft_ocr_one(base64_img)
    # res = formula_reg(base64_img)
    # print(res)