|
@@ -86,6 +86,12 @@ def image_process(image_np, image_path, is_from_pdf=False, is_from_docx=False, u
|
|
|
|
|
|
|
|
log("into image_preprocess")
|
|
log("into image_preprocess")
|
|
|
try:
|
|
try:
|
|
|
|
|
+ # 整体分辨率限制
|
|
|
|
|
+ if image_np.shape[0] > 2000 or image_np.shape[1] > 2000:
|
|
|
|
|
+ h, w = get_best_predict_size2(image_np, threshold=2000)
|
|
|
|
|
+ log("global image resize " + str(image_np.shape[:2]) + " -> " + str(h) + "," + str(w))
|
|
|
|
|
+ image_np = pil_resize(image_np, h, w)
|
|
|
|
|
+
|
|
|
# 图片倾斜校正，写入原来的图片路径
|
|
# 图片倾斜校正，写入原来的图片路径
|
|
|
# print("image_process", image_path)
|
|
# print("image_process", image_path)
|
|
|
g_r_i = get_rotated_image(image_np, image_path)
|
|
g_r_i = get_rotated_image(image_np, image_path)
|
|
@@ -217,7 +223,9 @@ def image_process(image_np, image_path, is_from_pdf=False, is_from_docx=False, u
|
|
|
# ocr图片过大内存溢出，需resize
|
|
# ocr图片过大内存溢出，需resize
|
|
|
start_time = time.time()
|
|
start_time = time.time()
|
|
|
threshold = 3000
|
|
threshold = 3000
|
|
|
|
|
+ ocr_resize_flag = 0
|
|
|
if image_np.shape[0] >= threshold or image_np.shape[1] >= threshold:
|
|
if image_np.shape[0] >= threshold or image_np.shape[1] >= threshold:
|
|
|
|
|
+ ocr_resize_flag = 1
|
|
|
best_h, best_w = get_best_predict_size2(image_np, threshold)
|
|
best_h, best_w = get_best_predict_size2(image_np, threshold)
|
|
|
# image_resize = cv2.resize(image_np, (best_w, best_h), interpolation=cv2.INTER_AREA)
|
|
# image_resize = cv2.resize(image_np, (best_w, best_h), interpolation=cv2.INTER_AREA)
|
|
|
image_resize = pil_resize(image_np, best_h, best_w)
|
|
image_resize = pil_resize(image_np, best_h, best_w)
|
|
@@ -246,7 +254,10 @@ def image_process(image_np, image_path, is_from_pdf=False, is_from_docx=False, u
|
|
|
# return []
|
|
# return []
|
|
|
|
|
|
|
|
# ocr resize后的bbox还原
|
|
# ocr resize后的bbox还原
|
|
|
- ratio = (image_np.shape[0]/best_h, image_np.shape[1]/best_w)
|
|
|
|
|
|
|
+ if ocr_resize_flag:
|
|
|
|
|
+ ratio = (image_np.shape[0]/best_h, image_np.shape[1]/best_w)
|
|
|
|
|
+ else:
|
|
|
|
|
+ ratio = (1, 1)
|
|
|
for i in range(len(bbox_list)):
|
|
for i in range(len(bbox_list)):
|
|
|
point = bbox_list[i]
|
|
point = bbox_list[i]
|
|
|
bbox_list[i] = [[int(point[0][0]*ratio[1]), int(point[0][1]*ratio[0])],
|
|
bbox_list[i] = [[int(point[0][0]*ratio[1]), int(point[0][1]*ratio[0])],
|
|
@@ -254,10 +265,6 @@ def image_process(image_np, image_path, is_from_pdf=False, is_from_docx=False, u
|
|
|
[int(point[2][0]*ratio[1]), int(point[2][1]*ratio[0])],
|
|
[int(point[2][0]*ratio[1]), int(point[2][1]*ratio[0])],
|
|
|
[int(point[3][0]*ratio[1]), int(point[3][1]*ratio[0])]]
|
|
[int(point[3][0]*ratio[1]), int(point[3][1]*ratio[0])]]
|
|
|
|
|
|
|
|
- # for _a,_b in zip(text_list,bbox_list):
|
|
|
|
|
- # print("bbox1",_a,_b)
|
|
|
|
|
-
|
|
|
|
|
-
|
|
|
|
|
# 调用现成方法形成表格
|
|
# 调用现成方法形成表格
|
|
|
try:
|
|
try:
|
|
|
from format_convert.convert_tree import TableLine
|
|
from format_convert.convert_tree import TableLine
|
