def text_recognize(_image_info, _box_info):
"""
文本识别
Args:
_image_info: 待识别的完整图像
_box_info: 图像中文本区域的位置
Returns: 文本区域位置的识别结果
"""
to_return_result = {'text': ''}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(
_image_info['bucket_name'],
_image_info['path']
)
cropped_image = get_rotated_box_roi_from_image(img, _box_info)
get_image_rotation = text_orientation_op.execute(cropped_image)
if get_image_rotation['orientation'] == TextImageOrientation.ORIENTATION_90:
rotated_image, _ = rotate_degree_img(cropped_image, 90)
elif get_image_rotation['orientation'] == TextImageOrientation.ORIENTATION_180:
rotated_image, _ = rotate_degree_img(cropped_image, 180)
elif get_image_rotation['orientation'] == TextImageOrientation.ORIENTATION_270:
rotated_image, _ = rotate_degree_img(cropped_image, 270)
else:
rotated_image = cropped_image
recognize_result = text_recognize_op.execute(rotated_image)
to_return_result['text'] = recognize_result['text']
return to_return_result
def ocr_result_visualization(_image_info, _box_info_list, _text_list):
"""
将检测的结果和识别的结果合并到一张图中
Args:
_image_info: 图片信息
_box_info_list: 所有box的列表
_text_list: 跟box顺序一致的识别结果
Returns: 合并后的图片的oss的路径
"""
to_return_result = {'bucket_name': '', 'path': ''}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
result_image = annotate_detect_rotated_bbox_and_text_result(
img, _box_info_list, _text_list, (0, 0, 255), 3)
date_string = get_date_string()
uuid_name = get_uuid_name()
image_path = os.path.join(date_string, uuid_name)
final_image_path = oss_handler.upload_image_file('result', image_path,
result_image, True, 50)
to_return_result['bucket_name'] = 'result'
to_return_result['path'] = final_image_path
return to_return_result
def text_detect(_image_info):
"""
文本检测
Args:
_image_info: 待检测的图像信息
Returns: 检测得到的所有box
"""
to_return_result = {'box_info': [], 'box_count': 0}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(
_image_info['bucket_name'],
_image_info['path']
)
if max(img.shape[:2]) > 1024:
candidate_img = resize_with_long_side(img, 1024)
else:
candidate_img = img
detect_result = db_res18_op.execute(candidate_img)
for m_box in detect_result['locations']:
m_box_info = m_box['box_info']
m_box_score = m_box['score']
to_return_result['box_info'].append({
'degree': m_box_info['degree'],
'center_x': m_box_info['center_x'],
'center_y': m_box_info['center_y'],
'box_height': m_box_info['box_height'],
'box_width': m_box_info['box_width'],
'score': m_box_score,
})
to_return_result['box_count'] = len(detect_result['locations'])
return to_return_result
def face_parsing(_image_info, _face_box_info, _face_landmark_info):
"""
人脸语义分区
Args:
_image_info: 待识别的完整图像
_face_box_info: 人脸所在区域
_face_landmark_info: 人脸landmark坐标信息
Returns: 人脸不同区域的mask的key
"""
to_return_result = {
'parsing_info': {
'bucket_name': '',
'path': ''
},
}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img,
_face_box_info,
_scale_ratio=1.5)
face_parsing_result = face_parsing_handler.execute(cropped_image,
_face_landmark_info)
parsing_result = face_parsing_result['semantic_segmentation']
date_string = get_date_string()
name_string = get_uuid_name()
target_path = os.path.join(date_string, name_string)
target_path = oss_handler.upload_numpy_array('intermediate', target_path,
parsing_result)
to_return_result['parsing_info']['bucket_name'] = 'intermediate'
to_return_result['parsing_info']['path'] = target_path
return to_return_result
def download_image_from_url(_image_url):
oss_helper = get_oss_handler()
download_result = image_download_op.execute(_image_url,
oss_helper,
_image_size_threshold=None)
return {
'image_info': {
'bucket_name': download_result['bucket_name'],
'path': download_result['saved_path'],
'height': download_result['image_height'],
'width': download_result['image_width'],
'channel': download_result['image_channel'],
},
}
def face_detect(_image_info):
"""
人脸检测
Args:
_image_info: 待识别的完整图像
Returns: 检测到的人脸区域
"""
to_return_result = {'face_count': 0, 'face_box_info': []}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
face_detect_result = face_detect_op.execute(img)
to_return_result['face_count'] = len(face_detect_result['locations'])
to_return_result['face_box_info'] = face_detect_result['locations']
return to_return_result
def text_recognize(_image_info, _box_info):
"""
文本识别
Args:
_image_info: 待识别的完整图像
_box_info: 图像中文本区域的位置
Returns: 文本区域位置的识别结果
"""
to_return_result = {'text': ''}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img, _box_info)
recognize_result = text_recognize_op.execute(cropped_image)
to_return_result['text'] = recognize_result['text']
return to_return_result
def face_liveness_detect(_image_info, _face_box_info):
"""
静默人脸活体检测
Args:
_image_info: 待识别的完整图像
_face_box_info: 人脸所在区域
Returns: 人脸的特征向量
"""
to_return_result = {"is_fake": False}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img, _face_box_info, 2.7)
liveness_result = mini_fasnetv2_handler.execute(cropped_image)
score = liveness_result['classification_scores']
to_return_result['is_fake'] = (score[1] > score[0]) and (score[1] >
score[2])
return to_return_result
def face_embedding(_image_info, _face_box_info, _face_landmark_info):
"""
人脸特征向量提取
Args:
_image_info: 待识别的完整图像
_face_box_info: 人脸所在区域
_face_landmark_info: 人脸landmark坐标信息
Returns: 人脸的特征向量
"""
to_return_result = {"face_feature_vector": []}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img, _face_box_info, 1.35)
embedding_result = asia_face_embedding_handler.execute(
cropped_image, _face_landmark_info)
to_return_result['face_feature_vector'] = embedding_result[
'feature_vector']
return to_return_result
def face_landmark(_image_info, _face_box_info):
"""
人脸landmark检测
Args:
_image_info: 待识别的完整图像
_face_box_info: 人脸所在区域
Returns: 人脸landmark坐标
"""
to_return_result = {
'points_count': 106,
'x_locations': [0] * 106,
'y_locations': [0] * 106,
}
oss_handler = get_oss_handler()
img = oss_handler.download_image_file(_image_info['bucket_name'],
_image_info['path'])
cropped_image = get_rotated_box_roi_from_image(img, _face_box_info, 1.35)
landmark_detect_result = landmark106p_detect_handler.execute(cropped_image)
to_return_result = landmark_detect_result.copy()
return to_return_result