|
1
|
|
|
# pylint: disable=missing-module-docstring,invalid-name |
|
2
|
|
|
from typing import Optional |
|
3
|
|
|
from itertools import count |
|
4
|
|
|
from collections import OrderedDict as odict |
|
5
|
|
|
|
|
6
|
|
|
import click |
|
7
|
|
|
|
|
8
|
|
|
from ocrd import Processor, OcrdPageResult |
|
9
|
|
|
from ocrd.decorators import ocrd_cli_options, ocrd_cli_wrap_processor |
|
10
|
|
|
from ocrd_modelfactory import page_from_file |
|
11
|
|
|
from ocrd_models import OcrdPage |
|
12
|
|
|
from ocrd_models.ocrd_page import ( |
|
13
|
|
|
BorderType, |
|
14
|
|
|
CoordsType, |
|
15
|
|
|
ReadingOrderType, |
|
16
|
|
|
UnorderedGroupType, |
|
17
|
|
|
) |
|
18
|
|
|
from ocrd_utils import bbox_from_points |
|
19
|
|
|
|
|
20
|
|
|
_SEGTYPES = [ |
|
21
|
|
|
"NoiseRegion", |
|
22
|
|
|
"LineDrawingRegion", |
|
23
|
|
|
"AdvertRegion", |
|
24
|
|
|
"ImageRegion", |
|
25
|
|
|
"ChartRegion", |
|
26
|
|
|
"MusicRegion", |
|
27
|
|
|
"GraphicRegion", |
|
28
|
|
|
"UnknownRegion", |
|
29
|
|
|
"CustomRegion", |
|
30
|
|
|
"SeparatorRegion", |
|
31
|
|
|
"MathsRegion", |
|
32
|
|
|
"TextRegion", |
|
33
|
|
|
"MapRegion", |
|
34
|
|
|
"ChemRegion", |
|
35
|
|
|
"TableRegion", |
|
36
|
|
|
"TextLine", |
|
37
|
|
|
"Word", |
|
38
|
|
|
"Glyph" |
|
39
|
|
|
] |
|
40
|
|
|
|
|
41
|
|
|
|
|
42
|
|
|
def get_border_bbox(pcgts): |
|
43
|
|
|
if pcgts.Page.Border is None: |
|
44
|
|
|
return [0, 0, pcgts.Page.imageWidth, pcgts.Page.imageHeight] |
|
45
|
|
|
return bbox_from_points(pcgts.Page.Border.Coords.points) |
|
46
|
|
|
|
|
47
|
|
|
def rename_segments(pcgts, start=1): |
|
48
|
|
|
renamed = {} |
|
49
|
|
|
rodict = pcgts.Page.get_ReadingOrderGroups() |
|
50
|
|
|
# get everything that has an identifier |
|
51
|
|
|
nodes = pcgts.xpath("//*[@id]") |
|
52
|
|
|
# filter segments |
|
53
|
|
|
segments = [segment for segment in map(pcgts.revmap.get, nodes) |
|
54
|
|
|
# get PAGE objects from matching etree nodes |
|
55
|
|
|
# but allow only hierarchy segments |
|
56
|
|
|
if segment.__class__.__name__.replace('Type', '') in _SEGTYPES] |
|
57
|
|
|
# count segments and rename them |
|
58
|
|
|
# fixme: or perhaps better to have each segment type named and counted differently? |
|
59
|
|
|
num = 0 |
|
60
|
|
|
regions = [] |
|
61
|
|
|
for num, segment in zip(count(start=start), segments): |
|
62
|
|
|
segtype = segment.original_tagname_ |
|
63
|
|
|
#parent = segment.parent_object_ |
|
64
|
|
|
newname = "seg%011d" % num |
|
65
|
|
|
assert not segment.id in renamed |
|
66
|
|
|
if segtype.endswith('Region') and segment.id in rodict: |
|
67
|
|
|
# update reading order |
|
68
|
|
|
roelem = rodict[segment.id] |
|
69
|
|
|
roelem.regionRef = newname |
|
70
|
|
|
renamed[segment.id] = newname |
|
71
|
|
|
segment.id = newname |
|
72
|
|
|
return num |
|
73
|
|
|
|
|
74
|
|
|
class MergeProcessor(Processor): |
|
75
|
|
|
def process_page_pcgts(self, *input_pcgts: Optional[OcrdPage], page_id: Optional[str] = None) -> OcrdPageResult: |
|
76
|
|
|
""" |
|
77
|
|
|
Merge PAGE segment hierarchy elements from all input file groups. |
|
78
|
|
|
|
|
79
|
|
|
For each page, open and deserialise PAGE input files. Rename all elements |
|
80
|
|
|
of the segment hierarchy to new (clash-free) identifers. Redefine the |
|
81
|
|
|
`Border` coordinates as the convex hull of all input borders. Then add all |
|
82
|
|
|
regions from all input files, concatenating them into a single `ReadingOrder` |
|
83
|
|
|
in the order of input file groups. |
|
84
|
|
|
|
|
85
|
|
|
Produce a new PAGE output file by serialising the resulting hierarchy. |
|
86
|
|
|
""" |
|
87
|
|
|
actual_pcgts = list(filter(None, input_pcgts)) |
|
88
|
|
|
assert len(set(pcgts.Page.imageFilename for pcgts in actual_pcgts)) == 1, \ |
|
89
|
|
|
"input files must all reference the same @imageFilename" |
|
90
|
|
|
# create new PAGE for image |
|
91
|
|
|
result = OcrdPageResult(page_from_file(actual_pcgts[0].Page.imageFilename)) |
|
92
|
|
|
# unify Border |
|
93
|
|
|
borders = [get_border_bbox(pcgts) for pcgts in actual_pcgts] |
|
94
|
|
|
minx, miny, maxx, maxy = zip(*borders) |
|
95
|
|
|
minx = min(minx) |
|
96
|
|
|
miny = min(miny) |
|
97
|
|
|
maxx = max(maxx) |
|
98
|
|
|
maxy = max(maxy) |
|
99
|
|
|
result.pcgts.Page.set_Border( |
|
100
|
|
|
BorderType(CoordsType( |
|
101
|
|
|
points=f"{minx},{miny} {maxx},{miny} {maxx},{maxy} {minx},{maxy}"))) |
|
102
|
|
|
# rename all segments |
|
103
|
|
|
num = 1 |
|
104
|
|
|
for pcgts in actual_pcgts: |
|
105
|
|
|
num = rename_segments(pcgts, num) |
|
106
|
|
|
# concatenate all regions |
|
107
|
|
|
ug = UnorderedGroupType(id="merged") |
|
108
|
|
|
result.pcgts.Page.set_ReadingOrder(ReadingOrderType(UnorderedGroup=ug)) |
|
109
|
|
|
for pcgts in actual_pcgts: |
|
110
|
|
|
for region in pcgts.Page.get_AllRegions(): |
|
111
|
|
|
adder = getattr(result.pcgts.Page, 'add_' + region.original_tagname_) |
|
112
|
|
|
adder(region) |
|
113
|
|
|
if pcgts.Page.ReadingOrder: |
|
114
|
|
|
group = pcgts.Page.ReadingOrder.OrderedGroup or pcgts.Page.ReadingOrder.UnorderedGroup |
|
115
|
|
|
adder = getattr(ug, 'add_' + group.original_tagname_) |
|
116
|
|
|
adder(group) |
|
117
|
|
|
return result |
|
118
|
|
|
|
|
119
|
|
|
@property |
|
120
|
|
|
def metadata_filename(self): |
|
121
|
|
|
return 'processor/builtin/dummy/ocrd-tool.json' |
|
122
|
|
|
|
|
123
|
|
|
@property |
|
124
|
|
|
def executable(self): |
|
125
|
|
|
return 'ocrd-merge' |
|
126
|
|
|
|
|
127
|
|
|
|
|
128
|
|
|
@click.command() |
|
129
|
|
|
@ocrd_cli_options |
|
130
|
|
|
def cli(*args, **kwargs): |
|
131
|
|
|
return ocrd_cli_wrap_processor(MergeProcessor, *args, **kwargs) |
|
132
|
|
|
|
OCR-D /
core
