283375/arcaea-offline-ocr
1
0
Fork 0
mirror of https://github.com/283375/arcaea-offline-ocr.git synced 2025-04-11 09:30:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

impr: minor improvements

Browse Source
This commit is contained in:
283375 2023-10-12 01:36:57 +08:00
parent 5e0642c832
commit 02599780e3
Signed by: 283375
SSH Key Fingerprint: SHA256:UcX0qg6ZOSDOeieKPGokA5h7soykG61nz2uxuQgVLSk
3 changed files with 36 additions and 39 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/arcaea_offline_ocr/device/common.py
View File

@ -13,6 +13,6 @@ class DeviceOcrResult:
max_recall: Optional[int] = None
song_id: Optional[str] = None
song_id_possibility: Optional[float] = None
clear_status: Optional[str] = None
clear_status: Optional[int] = None
partner_id: Optional[str] = None
partner_id_possibility: Optional[float] = None

3
src/arcaea_offline_ocr/device/ocr.py
View File

@ -47,8 +47,7 @@ class DeviceOcr:
continue
roi_ocr = cv2.fillPoly(roi_ocr, [contour], [0])
digit_rois = [
resize_fill_square(crop_xywh(roi_ocr, r), 20)
for r in sorted(filtered_rects, key=lambda r: r[0])
resize_fill_square(crop_xywh(roi_ocr, r), 20) for r in filtered_rects
]
samples = preprocess_hog(digit_rois)

70
src/arcaea_offline_ocr/ocr.py
View File

@ -1,5 +1,4 @@
import math
from copy import deepcopy
from typing import Optional, Sequence, Tuple
import cv2
@ -64,8 +63,7 @@ class FixRects:
new_h = new_bottom - new_y
new_rects.append((new_x, new_y, new_w, new_h))
return_rects = deepcopy(rects)
return_rects = [r for r in return_rects if r not in consumed_rects]
return_rects = [r for r in rects if r not in consumed_rects]
return_rects.extend(new_rects)
return return_rects
@ -80,42 +78,42 @@ class FixRects:
new_rects = []
for rect in rects:
rx, ry, rw, rh = rect
if rw / rh > rect_wh_ratio:
# consider this is a connected contour
connected_rects.append(rect)
if rw / rh <= rect_wh_ratio:
continue
# find the thinnest part
border_ignore = round(rw * width_range_ratio)
img_cropped = crop_xywh(
img_masked,
(border_ignore, ry, rw - border_ignore, rh),
)
white_pixels = {} # dict[x, white_pixel_number]
for i in range(img_cropped.shape[1]):
col = img_cropped[:, i]
white_pixels[rx + border_ignore + i] = np.count_nonzero(col > 200)
least_white_pixels = min(v for v in white_pixels.values() if v > 0)
x_values = [
x
for x, pixel in white_pixels.items()
if pixel == least_white_pixels
]
# select only middle values
x_mean = np.mean(x_values)
x_std = np.std(x_values)
x_values = [
x
for x in x_values
if x_mean - x_std * 1.5 <= x <= x_mean + x_std * 1.5
]
x_mid = round(np.median(x_values))
connected_rects.append(rect)
# split the rect
new_rects.extend(
[(rx, ry, x_mid - rx, rh), (x_mid, ry, rx + rw - x_mid, rh)]
)
# find the thinnest part
border_ignore = round(rw * width_range_ratio)
img_cropped = crop_xywh(
img_masked,
(border_ignore, ry, rw - border_ignore, rh),
)
white_pixels = {} # dict[x, white_pixel_number]
for i in range(img_cropped.shape[1]):
col = img_cropped[:, i]
white_pixels[rx + border_ignore + i] = np.count_nonzero(col > 200)
if all(v == 0 for v in white_pixels.values()):
return rects
least_white_pixels = min(v for v in white_pixels.values() if v > 0)
x_values = [
x for x, pixel in white_pixels.items() if pixel == least_white_pixels
]
# select only middle values
x_mean = np.mean(x_values)
x_std = np.std(x_values)
x_values = [
x for x in x_values if x_mean - x_std * 1.5 <= x <= x_mean + x_std * 1.5
]
x_mid = round(np.median(x_values))
# split the rect
new_rects.extend(
[(rx, ry, x_mid - rx, rh), (x_mid, ry, rx + rw - x_mid, rh)]
)
return_rects = deepcopy(rects)
return_rects = [r for r in rects if r not in connected_rects]
return_rects.extend(new_rects)
return return_rects