352 lines
13 KiB
Python
352 lines
13 KiB
Python
import re
|
||
import json
|
||
import cv2
|
||
import numpy as np
|
||
from PIL import Image, ImageDraw, ImageFont
|
||
import os
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# CONFIG
|
||
# ─────────────────────────────────────────────
|
||
INPUT_IMAGE = "page.png"
|
||
OUTPUT_IMAGE = "page_translated.png"
|
||
TRANSLATIONS_FILE = "output.txt"
|
||
BUBBLES_FILE = "bubbles.json"
|
||
FONT_PATH = "font.ttf"
|
||
FONT_FALLBACK = "/System/Library/Fonts/Helvetica.ttc"
|
||
FONT_COLOR = (0, 0, 0)
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# PARSE output.txt
|
||
# ─────────────────────────────────────────────
|
||
def parse_translations(filepath):
|
||
"""
|
||
Parses output.txt → {bubble_id: translated_text}.
|
||
Only bubbles present in the file are returned.
|
||
Absent IDs are left completely untouched on the page.
|
||
"""
|
||
translations = {}
|
||
with open(filepath, "r", encoding="utf-8") as f:
|
||
for line in f:
|
||
line = line.rstrip("\n")
|
||
if not re.match(r"^\s*#\d+", line):
|
||
continue
|
||
parts = re.split(r" {2,}", line.strip())
|
||
if len(parts) < 3:
|
||
continue
|
||
bubble_id = int(re.sub(r"[^0-9]", "", parts[0]))
|
||
translated = parts[-1].strip()
|
||
if translated.startswith("["):
|
||
continue
|
||
translations[bubble_id] = translated
|
||
|
||
print(f" ✅ {len(translations)} bubble(s) to translate: "
|
||
f"{sorted(translations.keys())}")
|
||
for bid, text in sorted(translations.items()):
|
||
print(f" #{bid}: {text}")
|
||
return translations
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# LOAD bubbles.json
|
||
# ─────────────────────────────────────────────
|
||
def load_bubble_boxes(filepath):
|
||
with open(filepath, "r", encoding="utf-8") as f:
|
||
raw = json.load(f)
|
||
boxes = {int(k): v for k, v in raw.items()}
|
||
print(f" ✅ Loaded {len(boxes)} bubble(s)")
|
||
for bid, val in sorted(boxes.items()):
|
||
print(f" #{bid}: ({val['x']},{val['y']}) "
|
||
f"{val['w']}×{val['h']}px")
|
||
return boxes
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# SAMPLE BACKGROUND COLOR
|
||
# ─────────────────────────────────────────────
|
||
def sample_bubble_background(cv_image, bubble_data):
|
||
"""
|
||
Samples the dominant background color inside the bbox
|
||
by averaging the brightest 10% of pixels.
|
||
Returns (B, G, R).
|
||
"""
|
||
x = max(0, bubble_data["x"])
|
||
y = max(0, bubble_data["y"])
|
||
x2 = min(cv_image.shape[1], x + bubble_data["w"])
|
||
y2 = min(cv_image.shape[0], y + bubble_data["h"])
|
||
|
||
region = cv_image[y:y2, x:x2]
|
||
if region.size == 0:
|
||
return (255, 255, 255)
|
||
|
||
gray = cv2.cvtColor(region, cv2.COLOR_BGR2GRAY)
|
||
threshold = np.percentile(gray, 90)
|
||
bg_mask = gray >= threshold
|
||
if not np.any(bg_mask):
|
||
return (255, 255, 255)
|
||
|
||
return tuple(int(c) for c in region[bg_mask].mean(axis=0))
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# ERASE ORIGINAL TEXT
|
||
# Fills the tight OCR bbox with the sampled
|
||
# background color. No extra expansion —
|
||
# the bbox from bubbles.json is already the
|
||
# exact size of the red squares.
|
||
# ─────────────────────────────────────────────
|
||
def erase_bubble_text(cv_image, bubble_data,
|
||
bg_color=(255, 255, 255)):
|
||
"""
|
||
Fills the bubble bounding box with bg_color.
|
||
|
||
Args:
|
||
cv_image : BGR numpy array (modified in place)
|
||
bubble_data : Dict with 'x','y','w','h'
|
||
bg_color : (B,G,R) fill color
|
||
"""
|
||
img_h, img_w = cv_image.shape[:2]
|
||
x = max(0, bubble_data["x"])
|
||
y = max(0, bubble_data["y"])
|
||
x2 = min(img_w, bubble_data["x"] + bubble_data["w"])
|
||
y2 = min(img_h, bubble_data["y"] + bubble_data["h"])
|
||
cv_image[y:y2, x:x2] = list(bg_color)
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# FIT FONT SIZE
|
||
# ─────────────────────────────────────────────
|
||
def fit_font_size(draw, text, max_w, max_h, font_path,
|
||
min_size=7, max_size=48):
|
||
"""
|
||
Finds the largest font size where word-wrapped text
|
||
fits inside (max_w × max_h).
|
||
"""
|
||
best_font = None
|
||
best_lines = [text]
|
||
|
||
for size in range(max_size, min_size - 1, -1):
|
||
try:
|
||
font = (ImageFont.truetype(font_path, size)
|
||
if font_path else ImageFont.load_default())
|
||
except Exception:
|
||
font = ImageFont.load_default()
|
||
|
||
words, lines, current = text.split(), [], ""
|
||
for word in words:
|
||
test = (current + " " + word).strip()
|
||
bb = draw.textbbox((0, 0), test, font=font)
|
||
if (bb[2] - bb[0]) <= max_w:
|
||
current = test
|
||
else:
|
||
if current:
|
||
lines.append(current)
|
||
current = word
|
||
if current:
|
||
lines.append(current)
|
||
|
||
lh = draw.textbbox((0, 0), "Ay", font=font)
|
||
line_h = (lh[3] - lh[1]) + 2
|
||
if line_h * len(lines) <= max_h:
|
||
best_font = font
|
||
best_lines = lines
|
||
break
|
||
|
||
return best_font or ImageFont.load_default(), best_lines
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# RENDER TEXT INTO BUBBLE
|
||
# ─────────────────────────────────────────────
|
||
def render_text_in_bubble(pil_image, bubble_data, text,
|
||
font_path, padding=8,
|
||
font_color=(0, 0, 0)):
|
||
"""
|
||
Renders translated text centered inside the tight bbox.
|
||
Font auto-sizes to fill the same w×h the original occupied.
|
||
"""
|
||
x, y = bubble_data["x"], bubble_data["y"]
|
||
w, h = bubble_data["w"], bubble_data["h"]
|
||
|
||
draw = ImageDraw.Draw(pil_image)
|
||
inner_w = max(1, w - padding * 2)
|
||
inner_h = max(1, h - padding * 2)
|
||
|
||
font, lines = fit_font_size(draw, text, inner_w, inner_h,
|
||
font_path)
|
||
|
||
lh_bb = draw.textbbox((0, 0), "Ay", font=font)
|
||
line_h = (lh_bb[3] - lh_bb[1]) + 2
|
||
total_h = line_h * len(lines)
|
||
start_y = y + padding + max(0, (inner_h - total_h) // 2)
|
||
|
||
for line in lines:
|
||
lb = draw.textbbox((0, 0), line, font=font)
|
||
line_w = lb[2] - lb[0]
|
||
start_x = x + padding + max(0, (inner_w - line_w) // 2)
|
||
draw.text((start_x, start_y), line,
|
||
font=font, fill=font_color)
|
||
start_y += line_h
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# RESOLVE FONT
|
||
# ─────────────────────────────────────────────
|
||
def resolve_font(font_path, fallback):
|
||
if font_path and os.path.exists(font_path):
|
||
print(f" ✅ Using font: {font_path}")
|
||
return font_path
|
||
if fallback and os.path.exists(fallback):
|
||
print(f" ⚠️ Fallback: {fallback}")
|
||
return fallback
|
||
print(" ⚠️ No font found. Using PIL default.")
|
||
return None
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# MAIN RENDERER
|
||
# ─────────────────────────────────────────────
|
||
def render_translated_page(
|
||
input_image = INPUT_IMAGE,
|
||
output_image = OUTPUT_IMAGE,
|
||
translations_file = TRANSLATIONS_FILE,
|
||
bubbles_file = BUBBLES_FILE,
|
||
font_path = FONT_PATH,
|
||
font_fallback = FONT_FALLBACK,
|
||
font_color = FONT_COLOR,
|
||
text_padding = 8,
|
||
debug = False,
|
||
):
|
||
"""
|
||
Pipeline:
|
||
1. Parse translations (only present IDs processed)
|
||
2. Load bubble boxes from bubbles.json
|
||
3. Cross-check IDs — absent ones left untouched
|
||
4. Sample background color per bubble
|
||
5. Erase original text (fill tight bbox)
|
||
6. Render translated text sized to fit the bbox
|
||
7. Save output
|
||
"""
|
||
print("=" * 55)
|
||
print(" MANGA TRANSLATOR — RENDERER")
|
||
print("=" * 55)
|
||
|
||
print("\n📄 Parsing translations...")
|
||
translations = parse_translations(translations_file)
|
||
if not translations:
|
||
print("❌ No translations found. Aborting.")
|
||
return
|
||
|
||
print(f"\n📦 Loading bubble data...")
|
||
bubble_boxes = load_bubble_boxes(bubbles_file)
|
||
if not bubble_boxes:
|
||
print("❌ No bubble data. Re-run manga-translator.py.")
|
||
return
|
||
|
||
translate_ids = set(translations.keys())
|
||
box_ids = set(bubble_boxes.keys())
|
||
to_process = sorted(translate_ids & box_ids)
|
||
untouched = sorted(box_ids - translate_ids)
|
||
missing = sorted(translate_ids - box_ids)
|
||
|
||
print(f"\n🔗 To process : {to_process}")
|
||
print(f" Untouched : {untouched}")
|
||
if missing:
|
||
print(f" ⚠️ In output.txt but no box: {missing}")
|
||
|
||
if not to_process:
|
||
print("❌ No matching IDs. Aborting.")
|
||
return
|
||
|
||
print(f"\n🖼️ Loading: {input_image}")
|
||
cv_image = cv2.imread(input_image)
|
||
if cv_image is None:
|
||
print(f"❌ Could not load: {input_image}")
|
||
return
|
||
print(f" {cv_image.shape[1]}×{cv_image.shape[0]}px")
|
||
|
||
# Sample backgrounds BEFORE erasing
|
||
print("\n🎨 Sampling backgrounds...")
|
||
bg_colors = {}
|
||
for bid in to_process:
|
||
bg_bgr = sample_bubble_background(
|
||
cv_image, bubble_boxes[bid])
|
||
bg_colors[bid] = bg_bgr
|
||
bg_rgb = (bg_bgr[2], bg_bgr[1], bg_bgr[0])
|
||
brightness = sum(bg_rgb) / 3
|
||
ink = "black" if brightness > 128 else "white"
|
||
print(f" #{bid}: RGB{bg_rgb} ink→{ink}")
|
||
|
||
# Erase
|
||
print("\n🧹 Erasing original text...")
|
||
for bid in to_process:
|
||
bd = bubble_boxes[bid]
|
||
erase_bubble_text(cv_image, bd, bg_color=bg_colors[bid])
|
||
print(f" ✅ #{bid} ({bd['w']}×{bd['h']}px)")
|
||
|
||
pil_image = Image.fromarray(
|
||
cv2.cvtColor(cv_image, cv2.COLOR_BGR2RGB))
|
||
|
||
print("\n🔤 Resolving font...")
|
||
resolved_font = resolve_font(font_path, font_fallback)
|
||
|
||
# Render
|
||
print("\n✍️ Rendering...")
|
||
for bid in to_process:
|
||
text = translations[bid]
|
||
bd = bubble_boxes[bid]
|
||
bg_rgb = (bg_colors[bid][2],
|
||
bg_colors[bid][1],
|
||
bg_colors[bid][0])
|
||
brightness = sum(bg_rgb) / 3
|
||
txt_color = (0, 0, 0) if brightness > 128 \
|
||
else (255, 255, 255)
|
||
|
||
render_text_in_bubble(
|
||
pil_image, bd, text,
|
||
font_path = resolved_font,
|
||
padding = text_padding,
|
||
font_color = txt_color,
|
||
)
|
||
print(f" ✅ #{bid}: '{text}' "
|
||
f"({bd['x']},{bd['y']}) {bd['w']}×{bd['h']}px")
|
||
|
||
if debug:
|
||
dbg = pil_image.copy()
|
||
dbg_draw = ImageDraw.Draw(dbg)
|
||
for bid, bd in sorted(bubble_boxes.items()):
|
||
color = (0, 200, 0) if bid in translate_ids \
|
||
else (160, 160, 160)
|
||
dbg_draw.rectangle(
|
||
[bd["x"], bd["y"],
|
||
bd["x"] + bd["w"], bd["y"] + bd["h"]],
|
||
outline=color, width=2)
|
||
dbg_draw.text((bd["x"] + 3, bd["y"] + 3),
|
||
f"#{bid}", fill=color)
|
||
dbg.save("debug_render.png")
|
||
print("\n 🐛 debug_render.png saved "
|
||
"(green=translated, grey=untouched)")
|
||
|
||
print(f"\n💾 Saving → {output_image}")
|
||
pil_image.save(output_image, "PNG")
|
||
print(" ✅ Done!")
|
||
print("=" * 55)
|
||
|
||
|
||
# ─────────────────────────────────────────────
|
||
# ENTRY POINT
|
||
# ─────────────────────────────────────────────
|
||
if __name__ == "__main__":
|
||
render_translated_page(
|
||
input_image = "page.png",
|
||
output_image = "page_translated.png",
|
||
translations_file = "output.txt",
|
||
bubbles_file = "bubbles.json",
|
||
font_path = "font.ttf",
|
||
font_fallback = "/System/Library/Fonts/Helvetica.ttc",
|
||
font_color = (0, 0, 0),
|
||
text_padding = 8,
|
||
debug = True,
|
||
) |