【校园AI Day-AI workshop】快速用PP-OCRv3识别并重命名

! mkdir dataset
! unzip -d dataset -q /home/aistudio/data/data149936/Scan_0012_0004.zip

import matplotlib.pyplot as pltfrom PIL import Image
%pylab inline

img = Image.open("/home/aistudio/dataset/Scan_0012_0004/Scan_0012_0019.jpg")
figsize = (48,24)
plt.figure("preview", figsize=figsize)
plt.imshow(img)
plt.show()

# 如仍需安装or安装更新，可以执行以下步骤# ! git clone https://github.com/PaddlePaddle/PaddleOCR.git -b dygra# 如果访问不了，可从国内尽享下载# ! git clone https://gitee.com/PaddlePaddle/PaddleOCR# 如果限网速慢，这里提前下载了一个，解压即可.! unzip -q PaddleOCR.zip

# 安装依赖包%cd /home/aistudio/PaddleOCR
! pip install -r requirements.txt
! pip install yacs gnureadline paddlenlp==2.2.1

%cd /home/aistudio/PaddleOCR
! mkdir inference# 下载超轻量级中文OCR模型的检测模型并解压! cd inference && wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar && tar xf ch_PP-OCRv3_det_infer.tar && rm ch_PP-OCRv3_det_infer.tar# 下载超轻量级中文OCR模型的识别模型并解压! cd inference && wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar && tar xf ch_PP-OCRv3_rec_infer.tar && rm ch_PP-OCRv3_rec_infer.tar# 下载超轻量级中文OCR模型的文本方向分类器模型并解压! cd inference && wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar && tar xf ch_ppocr_mobile_v2.0_cls_infer.tar && rm ch_ppocr_mobile_v2.0_cls_infer.tar

#  直接使用预训练的推理模型预测%cd /home/aistudio/PaddleOCR
!python tools/infer/predict_system.py \
    --image_dir="/home/aistudio/dataset/Scan_0012_0004/" \
    --det_model_dir="./inference/ch_PP-OCRv3_det_infer/" \
    --rec_model_dir="./inference/ch_PP-OCRv3_rec_infer/" \
    --cls_model_dir="./inference/ch_ppocr_mobile_v2.0_cls_infer"

# 查看预测后图片%cd /home/aistudio/PaddleOCR
!ls  ./inference_results/

%cd /home/aistudio/PaddleOCRimport matplotlib.pyplot as pltfrom PIL import Image
%matplotlib inline
img = Image.open("./inference_results/Scan_0012_0019.jpg")
plt.figure("preview", figsize=(48,24))
plt.imshow(img)
plt.show()

import jsonfrom PIL import Imageimport re"""
解析OCR预测结果，找出“铁塔名称”对应值
"""def parse_ocr_result(dataset_path,question_label_name="铁塔名称"):
    file = open(dataset_path+'/system_results.txt')
    
    newinfo = {}
    i = 0
    lines = ""
    while True:
        line = file.readline()        if not line:            break
        image_path,ocr_info = line.split("\t");
        ocr_infos = json.loads(ocr_info)
        question_label_points = [];        # 1.定位“铁塔名称”文字的位置
        for ocr_info in ocr_infos:            if ocr_info['transcription'] == question_label_name:
                question_label_points = ocr_info['points'] # 【左上，右上，右下，左下】
            
        if question_label_points == []:            continue
        # 2.找到“铁塔名称”文字右边的有效标签，可能分多行,多行文本按标签排序
        valid_labels=[];        for ocr_info in ocr_infos:
            label_points = ocr_info['points']            if ocr_info['transcription'] == question_label_name:                continue
            if label_points[0][0] < question_label_points[1][0]:                continue 
            if label_points[3][1] <  question_label_points[1][1]:                continue 
            if label_points[0][1] >  question_label_points[3][1]:                continue

            # lines += image_path +"\t" + ocr_info['transcription'] + "\n"
            #按Y轴从小到大的顺序添加
            if valid_labels == []:
                valid_labels.append(ocr_info) 
            else:                if valid_labels[-1]['points'][0][1] < label_points[0][1]:
                    valid_labels.append(ocr_info) 
                elif valid_labels[0]['points'][0][1] > label_points[0][1]:
                    valid_labels.insert(ocr_info)

        answer_text = ''
        for valid_label  in valid_labels:
            answer_text += valid_label['transcription']        # lines += image_path +"\t" + answer_text + "\n"
        lines += "mv "+image_path+" "+answer_text+".jpg \n"

    print(lines)    with open(dataset_path+'/rename_info.txt','w+',encoding='utf-8') as f2:
        f2.writelines(lines)

parse_ocr_result('/home/aistudio/PaddleOCR/inference_results', "铁塔名称")

import osimport jsonfrom PIL import Imageimport reimport mathdef get_images_question(dataset_path,question_label_name):
    file = open(dataset_path+'/system_results.txt')
    newinfo = {}
    i = 0
    lines = ""
    question_labels = [];    while True:
        line = file.readline()        if not line:            break
        image_path,ocr_info = line.split("\t");
        ocr_infos = json.loads(ocr_info)        # 1.定位“铁塔名称”文字的位置
        for ocr_info in ocr_infos:            if ocr_info['transcription'] == question_label_name:
                img = Image.open(dataset_path+'/'+image_path)
                ocr_info['width'] = img.width
                question_labels.append(ocr_info)    return question_labelsdef get_nearly_point(dataset_path,image_path,valid_questions):
    img = Image.open(dataset_path+'/'+image_path)
    
    width_nearly=[]    #保存图片宽度
    for question in valid_questions:
        width_nearly.append([question['transcription'],abs(question['width'] - img.width),question['points']])
    
    min_diff = width_nearly[0][1]
    min_points = width_nearly[0][2]    for item in width_nearly:        if item[0] == image_path: #排除自身
            continue
        if int(item[1]) < int(min_diff):
            min_diff = int(item[1])
            min_points = item[2]        
    return min_points"""
解析OCR预测结果，找出“铁塔名称”对应值
"""def parse_ocr_result2(dataset_path,valid_questions,question_label_name="铁塔名称"):
    file = open(dataset_path+'/system_results.txt')
    newinfo = {}
    i = 0
    lines = ""
    rename_info = []    while True:
        line = file.readline()        if not line:            break
        image_path,ocr_info = line.split("\t");
        ocr_infos = json.loads(ocr_info)
        question_label_points = [];        # 1.定位“铁塔名称”文字的位置
        for ocr_info in ocr_infos:            if ocr_info['transcription'] == question_label_name:
                question_label_points = ocr_info['points'] # 【左上，右上，右下，左下】
                
          
        if question_label_points == []:
            question_label_points = get_nearly_point(dataset_path,image_path,valid_questions)            # continue

        # 2.找到“铁塔名称”文字右边的有效标签，可能分多行,多行文本按标签排序
        valid_labels=[];        for ocr_info in ocr_infos:
            label_points = ocr_info['points']            if ocr_info['transcription'] == question_label_name:                continue
            if label_points[0][0] < question_label_points[1][0]:                continue 
            if label_points[3][1] <  question_label_points[1][1]:                continue 
            if label_points[0][1] >  question_label_points[3][1]:                continue

            # lines += image_path +"\t" + ocr_info['transcription'] + "\n"
            #按Y轴从小到大的顺序添加
            if valid_labels == []:
                valid_labels.append(ocr_info) 
            else:                if valid_labels[-1]['points'][0][1] < label_points[0][1]:
                    valid_labels.append(ocr_info) 
                elif valid_labels[0]['points'][0][1] > label_points[0][1]:
                    valid_labels.insert(ocr_info)        #  # 没识别出标签
        # if valid_labels == []:
        #     print(image_path,"图中未识别到 ",question_label_name," 标签对应的答案")
        
        answer_text = ''
        for valid_label  in valid_labels:
            answer_text += valid_label['transcription']        #防止重名
        i=1
        if answer_text in rename_info:
            answer_text  = answer_text+"（"+ str(i) +"）"
            i=i+1
        
        rename_info.append(answer_text)
        ext = os.path.splitext(image_path)[-1]
        lines += "mv "+image_path+" "+answer_text+ ext+" \n"
    print(lines)    with open(dataset_path+'/rename_info.sh','w+',encoding='utf-8') as f2:
        f2.writelines(lines)

valid_questions = get_images_question('/home/aistudio/PaddleOCR/inference_results',"铁塔名称")# print(valid_questions)parse_ocr_result2('/home/aistudio/PaddleOCR/inference_results',valid_questions, "铁塔名称")

# 重命名图片%cd /home/aistudio/dataset/Scan_0012_0004
! cp /home/aistudio/PaddleOCR/inference_results/rename_info.sh ./rename_info.sh
! chmod +x ./rename_info.sh
! ./rename_info.sh
! ls -l