Unverified Commit f9310954 authored by myhloli's avatar myhloli Committed by GitHub
Browse files

Merge pull request #4 from myhloli/dev-in-line-bbox

Dev in line bbox
parents 32fd7f95 64d67b5c
...@@ -23,10 +23,9 @@ from magic_pdf.pre_proc.ocr_cut_image import cut_image_and_table ...@@ -23,10 +23,9 @@ from magic_pdf.pre_proc.ocr_cut_image import cut_image_and_table
from magic_pdf.pre_proc.ocr_detect_layout import layout_detect from magic_pdf.pre_proc.ocr_detect_layout import layout_detect
from magic_pdf.pre_proc.ocr_dict_merge import ( from magic_pdf.pre_proc.ocr_dict_merge import (
merge_spans_to_line_by_layout, merge_spans_to_line_by_layout,
modify_y_axis
) )
from magic_pdf.pre_proc.ocr_span_list_modify import remove_spans_by_bboxes, remove_overlaps_min_spans, \ from magic_pdf.pre_proc.ocr_span_list_modify import remove_spans_by_bboxes, remove_overlaps_min_spans, \
adjust_bbox_for_standalone_block adjust_bbox_for_standalone_block,modify_y_axis,modify_inline_equation
from magic_pdf.pre_proc.remove_bbox_overlap import remove_overlap_between_bbox from magic_pdf.pre_proc.remove_bbox_overlap import remove_overlap_between_bbox
...@@ -184,8 +183,11 @@ def parse_pdf_by_ocr( ...@@ -184,8 +183,11 @@ def parse_pdf_by_ocr(
spans = cut_image_and_table(spans, page, page_id, book_name, save_path) spans = cut_image_and_table(spans, page, page_id, book_name, save_path)
# 行内公式调整, 高度调整至与同行文字高度一致(优先左侧, 其次右侧) # 行内公式调整, 高度调整至与同行文字高度一致(优先左侧, 其次右侧)
displayed_list = []
text_inline_lines = []
modify_y_axis(spans, displayed_list, text_inline_lines)
# 模型识别错误的行间公式, type类型转换成行内公式 # 模型识别错误的行间公式, type类型转换成行内公式
spans = modify_y_axis(spans) spans = modify_inline_equation(spans, displayed_list, text_inline_lines)
# bbox去除粘连 # bbox去除粘连
spans = remove_overlap_between_bbox(spans) spans = remove_overlap_between_bbox(spans)
......
...@@ -81,101 +81,6 @@ def merge_spans_to_line_by_layout(spans, layout_bboxes): ...@@ -81,101 +81,6 @@ def merge_spans_to_line_by_layout(spans, layout_bboxes):
def modify_y_axis(spans: list):
inline_list = []
displayed_list = []
text_list = []
image_list = []
table_list = []
spans.sort(key=lambda span: span['bbox'][1])
lines = []
current_line = [spans[0]]
if spans[0]["type"] in ["displayed_equation", "image", "table"]:
displayed_list.append(spans[0])
line_first_y0 = spans[0]["bbox"][1]
line_first_y = spans[0]["bbox"][3]
#用于给行间公式搜索
text_inline_lines = []
for span in spans[1:]:
# if span.get("content","") == "78.":
# print("debug")
# 如果当前的span类型为"displayed_equation" 或者 当前行中已经有"displayed_equation"
# image和table类型,同上
if span['type'] in ["displayed_equation", "image", "table"] or any(
s['type'] in ["displayed_equation", "image", "table"] for s in current_line):
#传入
if span["type"] in ["displayed_equation", "image", "table"]:
displayed_list.append(span)
# 则开始新行
lines.append(current_line)
if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
current_line = [span]
line_first_y0 = span["bbox"][1]
line_first_y = span["bbox"][3]
continue
# 如果当前的span与当前行的最后一个span在y轴上重叠,则添加到当前行
if __is_overlaps_y_exceeds_threshold(span['bbox'], current_line[-1]['bbox']):
if span["bbox"][1] < line_first_y0:
line_first_y0 = span["bbox"][1]
if span["bbox"][3] > line_first_y:
line_first_y = span["bbox"][3]
current_line.append(span)
else:
# 否则,开始新行
lines.append(current_line)
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
current_line = [span]
line_first_y0 = span["bbox"][1]
line_first_y = span["bbox"][3]
# 添加最后一行
if current_line:
lines.append(current_line)
if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
for line in text_inline_lines:
# 按照x0坐标排序
current_line = line[0]
current_line.sort(key=lambda span: span['bbox'][0])
#调整每一个文字行内bbox统一
for line in text_inline_lines:
current_line, (line_first_y0, line_first_y) = line
for span in current_line:
span["bbox"][1] = line_first_y0
span["bbox"][3] = line_first_y
#错误行间公式转行内公式
j = 0
for i in range(len(displayed_list)):
# if i == 8:
# print("debug")
span = displayed_list[i]
span_y0, span_y = span["bbox"][1], span["bbox"][3]
while j < len(text_inline_lines):
text_line = text_inline_lines[j]
y0, y1 = text_line[1]
if (span_y0 < y0 and span_y > y0 or span_y0 < y1 and span_y > y1 or span_y0 < y0 and span_y > y1) and __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
span["bbox"][1] = y0
# span["bbox"][3] = y1
#调整公式类型
if span["type"] == "displayed_equation":
span["type"] = "inline_equation"
break
elif span_y < y0 or span_y0 < y0 and span_y > y0 and not __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
break
else:
j += 1
return spans
...@@ -43,3 +43,104 @@ def adjust_bbox_for_standalone_block(spans): ...@@ -43,3 +43,104 @@ def adjust_bbox_for_standalone_block(spans):
# 调整span的y0和span2的y0一致 # 调整span的y0和span2的y0一致
sb_span['bbox'][1] = text_span['bbox'][1] sb_span['bbox'][1] = text_span['bbox'][1]
return spans return spans
def modify_y_axis(spans: list, displayed_list: list, text_inline_lines: list):
# displayed_list = []
spans.sort(key=lambda span: span['bbox'][1])
lines = []
current_line = [spans[0]]
if spans[0]["type"] in ["displayed_equation", "image", "table"]:
displayed_list.append(spans[0])
line_first_y0 = spans[0]["bbox"][1]
line_first_y = spans[0]["bbox"][3]
#用于给行间公式搜索
# text_inline_lines = []
for span in spans[1:]:
# if span.get("content","") == "78.":
# print("debug")
# 如果当前的span类型为"displayed_equation" 或者 当前行中已经有"displayed_equation"
# image和table类型,同上
if span['type'] in ["displayed_equation", "image", "table"] or any(
s['type'] in ["displayed_equation", "image", "table"] for s in current_line):
#传入
if span["type"] in ["displayed_equation", "image", "table"]:
displayed_list.append(span)
# 则开始新行
lines.append(current_line)
if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
current_line = [span]
line_first_y0 = span["bbox"][1]
line_first_y = span["bbox"][3]
continue
# 如果当前的span与当前行的最后一个span在y轴上重叠,则添加到当前行
if __is_overlaps_y_exceeds_threshold(span['bbox'], current_line[-1]['bbox']):
if span["type"] == "text":
line_first_y0 = span["bbox"][1]
line_first_y = span["bbox"][3]
current_line.append(span)
else:
# 否则,开始新行
lines.append(current_line)
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
current_line = [span]
line_first_y0 = span["bbox"][1]
line_first_y = span["bbox"][3]
# 添加最后一行
if current_line:
lines.append(current_line)
if len(current_line) > 1 or current_line[0]["type"] in ["text", "inline_equation"]:
text_inline_lines.append((current_line, (line_first_y0, line_first_y)))
for line in text_inline_lines:
# 按照x0坐标排序
current_line = line[0]
current_line.sort(key=lambda span: span['bbox'][0])
#调整每一个文字行内bbox统一
for line in text_inline_lines:
current_line, (line_first_y0, line_first_y) = line
for span in current_line:
span["bbox"][1] = line_first_y0
span["bbox"][3] = line_first_y
# return spans, displayed_list, text_inline_lines
def modify_inline_equation(spans: list, displayed_list: list, text_inline_lines: list):
#错误行间公式转行内公式
j = 0
for i in range(len(displayed_list)):
# if i == 8:
# print("debug")
span = displayed_list[i]
span_y0, span_y = span["bbox"][1], span["bbox"][3]
while j < len(text_inline_lines):
text_line = text_inline_lines[j]
y0, y1 = text_line[1]
if (span_y0 < y0 and span_y > y0 or span_y0 < y1 and span_y > y1 or span_y0 < y0 and span_y > y1) and __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
span["bbox"][1] = y0
# span["bbox"][3] = y1
#调整公式类型
if span["type"] == "displayed_equation":
if j+1 >= len(text_inline_lines):
span["type"] = "inline_equation"
else:
y0_next, y1_next = text_inline_lines[j + 1][1]
if not __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0_next, 0, y1_next)):
span["type"] = "inline_equation"
break
elif span_y < y0 or span_y0 < y0 and span_y > y0 and not __is_overlaps_y_exceeds_threshold(span['bbox'], (0, y0, 0, y1)):
break
else:
j += 1
return spans
\ No newline at end of file
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment