def process(buffer):
logger.debug("从web读取数据len:%r", len(buffer))
if len(buffer) == 0: return False, "Image is null"
data_array = np.frombuffer(buffer, dtype=np.uint8)
image = cv2.imdecode(data_array, cv2.IMREAD_COLOR)
if image is None:
logger.error("图像解析失败") #有可能从字节数组解析成图片失败
return "图像角度探测失败"
logger.debug("从字节数组变成图像的shape:%r", image.shape)
global input_images, sess
_classes = pred.pred(sess, classes, input_images, np.array([image]))
return pred.CLASS_NAME[_classes[0]]
def process(buffer, image_name):
logger.debug("从web读取数据len:%r", len(buffer))
if len(buffer) == 0: return False, "Image is null"
#先给他转成ndarray(numpy的)
data_array = np.frombuffer(buffer, dtype=np.uint8)
#从ndarray中读取图片，有raw数据变成一个图片GBR数据
#出来的数据，其实就是有维度了，就是原图的尺寸，如160x70
image = cv2.imdecode(data_array, cv2.IMREAD_COLOR)[:, :, ::-1]
# cv2.imwrite("test1.png",image) BGR
# cv2.imwrite("test2.png",image[:,:,::-1]) RGB
if image is None:
logger.error("图像解析失败") #有可能从字节数组解析成图片失败
return False, None
logger.debug("从字节数组变成图像的shape:%r", image.shape)
result = ctpn.pred(sess_ctpn, [image], [image_name])
for r in result:
# 从opencv的np array格式，转成原始图像，再转成base64
r['image'] = ocr_utils.tobase64(r['image'])
# logger.debug("预测返回结果：%r",result[0])
small_images = ocr_utils.crop_small_images(image, result[0]['boxes'])
# small_images = small_images[:2] # 测试用，为了提高测试速度，只处理2个
all_txt = crnn.pred(small_images, sess_crnn)
# 小框们的文本们
result[0]['text'] = all_txt
# 小框们的图片的base64
result[0]['small_images'] = ocr_utils.tobase64(small_images)
logger.debug("最终的预测结果为：%r", all_txt)
# logger.debug("最终的预测的子图:%r",result[0]['small_images'])
return True, result[0]
def process(image, image_name="test.jpg", is_verbose=False):
# result:[{
# name: 'xxx.png',
# 'box': {
# [1, 1, 1, 1],
# [2, 2, 2, 2]
# },
# image : <draw image numpy array>,
# 'f1': 0.78
# }
# }, ]
#global crnn_sess,ctpn_sess
result = ctpn.pred(ctpn_sess, [image], [image_name], ctpn_graph)
# logger.debug("预测返回结果：%r",result[0])
small_images = ocr_utils.crop_small_images(image, result[0]['boxes'])
result[0]['text'] = process_crnn(small_images) # 小框们的文本们
# 仅仅调试CTPN
# all_txt = ['']*len(small_images)
# 返回原图和切出来的小图，这个是为了调试用
if is_verbose:
# 小框们的图片的base64
result[0]['small_images'] = ocr_utils.nparray2base64(small_images)
# 这个是为了，把图片再回显到网页上用
for r in result:
# 从opencv的np array格式，转成原始图像，再转成base64
if r.get('image', None) is not None:
logger.debug("返回的大图:%r", r['image'].shape)
r['image'] = ocr_utils.nparray2base64(r['image'])
else:
logger.debug("返回大图为空")
# logger.debug("最终的预测的子图:%r",result[0]['small_images'])
return True, result[0]
def main():
input_images, classes = pred.init_model()
sess = pred.restore_session()
image_name_list = pred.get_images()
pbar = tqdm(total=len(image_name_list))
i = 0
for image_name in image_name_list:
start = time.time()
logger.info("探测图片[%s]开始", image_name)
pbar.update(1)
try:
img = cv2.imread(image_name)
_classes = pred.pred(sess, classes, input_images, np.array([img]))
if _classes[0] != 0:
select_image(image_name, _classes[0])
except Exception as e:
traceback.print_exc()
logger.error("处理图片[%s]发生错误：%s", image_name, str(e))
continue
i += 1
logger.debug("处理完第%d张图片，耗时:%f", i, time.time() - start)
pbar.close()