3

实现内容：OCR (get word from image) API：VisionAPI(Google) 语言：Python3.x

Q. 所有响应（块/段落/单词/符号）都应该有信心（param）。当然块/段落/单词有信心，但符号很少没有信心。概率是（67个字符/ 48036个字符）=0.14% ......我无法理解这种现象......

CODE（确定上述概率）

import json
import glob
json_path_from_list=glob.glob('./RawJSON/*')
#number of characters not to have confidence
no_confidence_number=0
#number of all characters
char_number=0
for json_path_from in json_path_from_list:
    with open(json_path_from) as fr:
        json_data = json.load(fr)
        for response_j in json_data["responses"]:
            for page_j in response_j["fullTextAnnotation"]["pages"]:
                for block_j in page_j["blocks"]:
                    x=block_j["confidence"]
                    for paragraph_j in block_j["paragraphs"]:
                        x=paragraph_j["confidence"]
                        for word_j in paragraph_j["words"]:
                            x=word_j["confidence"]
                            #initialize (confidence list (character having confidence))
                            symbol_confidence_list=[]
                            #initialize (word including characters not having confidence)
                            word=""
                            #initialize (print flag)
                            flag=0
                            for symbol_j in word_j["symbols"]:
                                #count
                                char_number+=1
                                try:
                                    #whether character has confidence or not 
                                    x=symbol_j["confidence"]
                                    #word including characters not having confidence
                                    word=word+symbol_j["text"]
                                    #confidence list (character having confidence)
                                    symbol_confidence_list.append(symbol_j["confidence"])
                                except:
                                    #word including characters not having confidence
                                    word=word+symbol_j["text"]
                                    #character not having confidence
                                    no_confidence_text=symbol_j
                                    #print flag
                                    flag=1
                            #print flag=1
                            if flag==1:
                                #count
                                no_confidence_number+=1
                                print(json_path_from)
                                print(word)
                                print(no_confidence_text)
                                #Wrapper's confidence
                                print(word_j["confidence"])
                                print(paragraph_j["confidence"])
                                print(block_j["confidence"])
                                print(symbol_confidence_list)
                                print("____________________________＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿＿")
                            else:pass
print(no_confidence_number)
print(char_number)

为了确定，请检查代码（从 VisionAPI 获取响应，并保存 .json）

import json
import base64
import requests
#function to encode base64
def convert_img_to_base64(img_path):
    img_data = open(img_path, 'rb').read()
    img_base64 = base64.b64encode(img_data).decode()
    return img_base64
#OCR class
class OCR:
    def __init__(self,api_key):
        self.api_key=api_key
    def get_json_responce_from_img(self,img_base64,json_request):
        json_request['requests'][0]['image']['content']=img_base64
        url="https://vision.googleapis.com/v1/images:annotate?key="+self.api_key
        result=requests.post(url,json.dumps(json_request),headers={'Content-Type': 'application/json'})
        return result.text
#API_key
API_KEY="XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"
#image path
img_path="./test.jpg"
#JSON REQUEST
json_request={'requests':[{'image':{'content':""},'features': {'type': "DOCUMENT_TEXT_DETECTION"},'imageContext':{'languageHints':"ja"}}]}
img_base64=convert_img_to_base64(img_path)
#create instance OCR class
ocr=OCR(API_KEY)
#JSON RESPONSE
data=json.loads(ocr.get_json_responce_from_img(img_base64,json_request))
#save JSON RESPONSE as .json
with open("./test.json",'w') as fw:
    json.dump(data,fw,indent=4)
4

0 回答 0