AIFFEL 아이펠 40일차

문자를 읽을 수 있는 딥러닝

OCR(Optical Character Recognition, 광학 문자 인식)

과정)

1. Text Detection(문자 검출): 입력받은 사진 속에서 문자의 위치를 찾는다.

 - 방식1) Regression(회귀): 기준으로 하는 박스 대비 문자의 박스가 얼마나 차이나는지 학습한다.

 - 방식2) Segmentation(세그멘테이션): 픽셀 단위로 해당 픽셀이 문자를 표현하는지를 분류하는 문제(pixel-wise classification)

2. Text Recognition(문자 인식): 찾은 문자 영역으로부터 문자를 읽어낸다.

 - CRNN: 이미지 내의 문자 인식 모델의 기본적인 방법 중 하나. 이미지 내의 텍스트와 연관된 특징을 CNN을 통해 추출한 후, 스텝 단위의 문자 정보를 RNN으로 인식한다.

Ibrahim, Ahmed Sobhy Elnady. End-To-End Text Detection Using Deep Learning. Diss. Virginia Tech, 2017. (https://arxiv.org/abs/1507.05717)

https://tv.naver.com/v/4578167

글자읽는 AI: 밑바닥부터 외국어 정복까지

https://www.youtube.com/watch?v=ckRFBl_XWFg 

---

<keras-ocr 써보기>

 

import matplotlib.pyplot as plt
import keras_ocr

# keras-ocr이 detector과 recognizer를 위한 모델을 자동으로 다운로드받게 됩니다. 
pipeline = keras_ocr.pipeline.Pipeline()

# Plot the predictions
fig, axs = plt.subplots(nrows=len(images), figsize=(20, 20))
for idx, ax in enumerate(axs):
    keras_ocr.tools.drawAnnotations(image=images[idx], 
                                    predictions=prediction_groups[idx][0], ax=ax)

# Plot the predictions
fig, axs = plt.subplots(nrows=len(images), figsize=(20, 20))
for idx, ax in enumerate(axs):
    keras_ocr.tools.drawAnnotations(image=images[idx], 
                                    predictions=prediction_groups[idx][0], ax=ax)

---

<Tesseract OCR 써보기>

 

import os
import pytesseract
from PIL import Image
from pytesseract import Output
import matplotlib.pyplot as plt

# OCR Engine modes(–oem):
# 0 - Legacy engine only.
# 1 - Neural nets LSTM engine only.
# 2 - Legacy + LSTM engines.
# 3 - Default, based on what is available.

# Page segmentation modes(–psm):
# 0 - Orientation and script detection (OSD) only.
# 1 - Automatic page segmentation with OSD.
# 2 - Automatic page segmentation, but no OSD, or OCR.
# 3 - Fully automatic page segmentation, but no OSD. (Default)
# 4 - Assume a single column of text of variable sizes.
# 5 - Assume a single uniform block of vertically aligned text.
# 6 - Assume a single uniform block of text.
# 7 - Treat the image as a single text line.
# 8 - Treat the image as a single word.
# 9 - Treat the image as a single word in a circle.
# 10 - Treat the image as a single character.
# 11 - Sparse text. Find as much text as possible in no particular order.
# 12 - Sparse text with OSD.
# 13 - Raw line. Treat the image as a single text line, bypassing hacks that are Tesseract-specific.

def crop_word_regions(image_path='./images/sample.png', output_path='./output'):
    if not os.path.exists(output_path):
        os.mkdir(output_path)
    custom_oem_psm_config = r'--oem 3 --psm 3'
    image = Image.open(image_path)

    recognized_data = pytesseract.image_to_data(
        image, lang='eng',    # 한국어라면 lang='kor'
        config=custom_oem_psm_config,
        output_type=Output.DICT
    )
    
    top_level = max(recognized_data['level'])
    index = 0
    cropped_image_path_list = []
    for i in range(len(recognized_data['level'])):
        level = recognized_data['level'][i]
    
        if level == top_level:
            left = recognized_data['left'][i]
            top = recognized_data['top'][i]
            width = recognized_data['width'][i]
            height = recognized_data['height'][i]
            
            output_img_path = os.path.join(output_path, f"{str(index).zfill(4)}.png")
            print(output_img_path)
            cropped_image = image.crop((
                left,
                top,
                left+width,
                top+height
            ))
            cropped_image.save(output_img_path)
            cropped_image_path_list.append(output_img_path)
            index += 1
    return cropped_image_path_list


work_dir = os.getenv('HOME')+'/aiffel/ocr_python'
img_file_path = work_dir + '/test1.jpeg'   #테스트용 이미지 경로입니다. 본인이 선택한 파일명으로 바꿔주세요. 

cropped_image_path_list = crop_word_regions(img_file_path, work_dir)

def recognize_images(cropped_image_path_list):
    custom_oem_psm_config = r'--oem 3 --psm 7'
    
    for image_path in cropped_image_path_list:
        image = Image.open(image_path)
        recognized_data = pytesseract.image_to_string(
            image, lang='eng',    # 한국어라면 lang='kor'
            config=custom_oem_psm_config,
            output_type=Output.DICT
        )
        print(recognized_data['text'])
    print("Done")

# 위에서 준비한 문자 영역 파일들을 인식하여 얻어진 텍스트를 출력합니다.
recognize_images(cropped_image_path_list)