Added all

batch-clean.sh

@@ -0,0 +1,228 @@
+#!/usr/bin/env bash
+# ============================================================
+# batch-clean.sh
+# Batch manga text removal (inpainting) using bubbles.json
+#
+# Usage:
+#   ./batch-clean.sh <folder>
+#   ./batch-clean.sh <folder> --start 3 --end 7
+#
+# Output per page lands in:
+#   <folder>/translated/<page_stem>/
+#     └── <page_stem>_cleaned.png
+# ============================================================
+
+set -uo pipefail
+
+# ─────────────────────────────────────────────────────────────
+# CONFIGURATION
+# ─────────────────────────────────────────────────────────────
+START_PAGE=1
+END_PAGE=999999
+PYTHON_BIN="python"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+CLEANER="${SCRIPT_DIR}/clean_bubbles.py"
+
+# ─────────────────────────────────────────────────────────────
+# COLOURS
+# ─────────────────────────────────────────────────────────────
+RED='\033[0;31m'
+GREEN='\033[0;32m'
+YELLOW='\033[1;33m'
+CYAN='\033[0;36m'
+BOLD='\033[1m'
+RESET='\033[0m'
+
+# ─────────────────────────────────────────────────────────────
+# HELPERS
+# ─────────────────────────────────────────────────────────────
+usage() {
+    echo ""
+    echo -e "${BOLD}Usage:${RESET}"
+    echo "  $0 <folder> [options]"
+    echo ""
+    echo -e "${BOLD}Options:${RESET}"
+    echo "  --start         First page number     (default: 1)"
+    echo "  --end           Last  page number     (default: all)"
+    echo "  --python        Python binary         (default: python)"
+    echo "  --help,    -h   Show this help"
+    echo ""
+    echo -e "${BOLD}Examples:${RESET}"
+    echo "  $0 pages-for-tests"
+    echo "  $0 pages-for-tests --start 3 --end 7"
+    echo ""
+}
+
+log_info()    { echo -e "${CYAN}ℹ️  $*${RESET}"; }
+log_ok()      { echo -e "${GREEN}✅  $*${RESET}"; }
+log_warn()    { echo -e "${YELLOW}⚠️  $*${RESET}"; }
+log_error()   { echo -e "${RED}❌  $*${RESET}"; }
+log_section() {
+    echo -e "\n${BOLD}${CYAN}══════════════════════════════════════════${RESET}"
+    echo -e "${BOLD}${CYAN}  🧹  $*${RESET}"
+    echo -e "${BOLD}${CYAN}══════════════════════════════════════════${RESET}"
+}
+
+# ─────────────────────────────────────────────────────────────
+# ARGUMENT PARSING
+# ─────────────────────────────────────────────────────────────
+if [[ $# -eq 0 ]]; then
+    log_error "No folder specified."
+    usage
+    exit 1
+fi
+
+FOLDER="$1"
+shift
+
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        --start)      START_PAGE="$2";  shift 2 ;;
+        --end)        END_PAGE="$2";    shift 2 ;;
+        --python)     PYTHON_BIN="$2";  shift 2 ;;
+        --help|-h)    usage; exit 0 ;;
+        *)
+            log_error "Unknown option: $1"
+            usage
+            exit 1
+            ;;
+    esac
+done
+
+# ─────────────────────────────────────────────────────────────
+# VALIDATION
+# ─────────────────────────────────────────────────────────────
+if [[ ! -d "$FOLDER" ]]; then
+    log_error "Folder not found: $FOLDER"
+    exit 1
+fi
+
+if [[ ! -f "$CLEANER" ]]; then
+    log_error "clean_bubbles.py not found at: $CLEANER"
+    exit 1
+fi
+
+if ! command -v "$PYTHON_BIN" &>/dev/null; then
+    log_error "Python binary not found: $PYTHON_BIN"
+    log_error "Try --python python3"
+    exit 1
+fi
+
+# ─────────────────────────────────────────────────────────────
+# DISCOVER IMAGES
+# ─────────────────────────────────────────────────────────────
+ALL_IMAGES=()
+while IFS= read -r -d '' img; do
+    ALL_IMAGES+=("$img")
+done < <(
+    find "$FOLDER" -maxdepth 1 -type f \
+        \( -iname "*.jpg" -o -iname "*.jpeg" \
+           -o -iname "*.png" -o -iname "*.webp" \) \
+        -print0 | sort -z
+)
+
+TOTAL=${#ALL_IMAGES[@]}
+
+if [[ $TOTAL -eq 0 ]]; then
+    log_error "No image files found in: $FOLDER"
+    exit 1
+fi
+
+# ─────────────────────────────────────────────────────────────
+# SLICE TO REQUESTED PAGE RANGE
+# ─────────────────────────────────────────────────────────────
+PAGES=()
+for i in "${!ALL_IMAGES[@]}"; do
+    PAGE_NUM=$(( i + 1 ))
+    if [[ $PAGE_NUM -ge $START_PAGE && $PAGE_NUM -le $END_PAGE ]]; then
+        PAGES+=("${ALL_IMAGES[$i]}")
+    fi
+done
+
+if [[ ${#PAGES[@]} -eq 0 ]]; then
+    log_error "No pages in range [${START_PAGE}, ${END_PAGE}] (total: ${TOTAL})"
+    exit 1
+fi
+
+# ─────────────────────────────────────────────────────────────
+# SUMMARY HEADER
+# ─────────────────────────────────────────────────────────────
+log_section "BATCH MANGA CLEANER"
+log_info "📂  Folder   : $(realpath "$FOLDER")"
+log_info "📄  Pages    : ${#PAGES[@]} of ${TOTAL} total"
+log_info "🔢  Range    : ${START_PAGE} → ${END_PAGE}"
+echo ""
+
+# ─────────────────────────────────────────────────────────────
+# PROCESS EACH PAGE
+# ─────────────────────────────────────────────────────────────
+PASS=0
+FAIL=0
+FAIL_LIST=()
+
+for i in "${!PAGES[@]}"; do
+    IMAGE="${PAGES[$i]}"
+    PAGE_NUM=$(( START_PAGE + i ))
+    STEM="$(basename "${IMAGE%.*}")"
+    WORKDIR="${FOLDER}/translated/${STEM}"
+    
+    echo ""
+    echo -e "${BOLD}──────────────────────────────────────────${RESET}"
+    echo -e "${BOLD}  🖼️  [${PAGE_NUM}/${TOTAL}]  ${STEM}${RESET}"
+    echo -e "${BOLD}──────────────────────────────────────────${RESET}"
+
+    OUTPUT_JSON="${WORKDIR}/bubbles.json"
+    OUTPUT_CLEANED="${WORKDIR}/${STEM}_cleaned.png"
+
+    if [[ ! -f "$OUTPUT_JSON" ]]; then
+        log_warn "Skipping: bubbles.json not found in ${WORKDIR}"
+        FAIL=$(( FAIL + 1 ))
+        FAIL_LIST+=("${STEM} (No JSON)")
+        continue
+    fi
+
+    log_info "🗂️  Image : $(basename "$IMAGE")"
+    log_info "🧹  Cleaning text..."
+
+    # ── Run the cleaner ───────────────────────────────────────
+    if "$PYTHON_BIN" "$CLEANER" \
+            -i "$IMAGE" \
+            -j "$OUTPUT_JSON" \
+            -o "$OUTPUT_CLEANED"; then
+
+        if [[ -f "$OUTPUT_CLEANED" ]]; then
+            log_ok "Cleaned image saved → ${STEM}_cleaned.png"
+            PASS=$(( PASS + 1 ))
+        else
+            log_error "Script ran but output image is missing."
+            FAIL=$(( FAIL + 1 ))
+            FAIL_LIST+=("${STEM} (Missing Output)")
+        fi
+    else
+        log_error "Page ${PAGE_NUM} FAILED — check output above."
+        FAIL=$(( FAIL + 1 ))
+        FAIL_LIST+=("${STEM} (Script Error)")
+    fi
+
+done
+
+# ─────────────────────────────────────────────────────────────
+# FINAL SUMMARY
+# ─────────────────────────────────────────────────────────────
+log_section "BATCH CLEANING COMPLETE"
+echo -e "  ✅  ${GREEN}Passed : ${PASS}${RESET}"
+echo -e "  ❌  ${RED}Failed : ${FAIL}${RESET}"
+
+if [[ ${#FAIL_LIST[@]} -gt 0 ]]; then
+    echo ""
+    log_warn "Failed pages:"
+    for NAME in "${FAIL_LIST[@]}"; do
+        echo -e "    ❌  ${RED}${NAME}${RESET}"
+    done
+fi
+
+echo ""
+log_info "📦  Output folder: $(realpath "${FOLDER}/translated")"
+echo ""
+
+[[ $FAIL -eq 0 ]] && exit 0 || exit 1

clean_bubbles.py

@@ -0,0 +1,109 @@
+import cv2
+import numpy as np
+import json
+import os
+import argparse
+
+def clean_text_from_box(image_bgr, box, bg_type="complex"):
+    """
+    Erases the text inside a specific bounding box.
+    Uses a solid white fill for 'white' backgrounds, and inpainting for 'complex' ones.
+    """
+    x, y = int(box["x"]), int(box["y"])
+    w, h = int(box["w"]), int(box["h"])
+    
+    # Image dimensions for boundary checking
+    img_h, img_w = image_bgr.shape[:2]
+    
+    # Ensure coordinates don't go outside the image boundaries
+    x = max(0, x)
+    y = max(0, y)
+    w = min(w, img_w - x)
+    h = min(h, img_h - y)
+    
+    # 1. Fast Path: Plain white background
+    if bg_type == "white":
+        cv2.rectangle(image_bgr, (x, y), (x+w, y+h), (255, 255, 255), -1)
+        return image_bgr
+        
+    # 2. Complex Path: Inpainting for screentones/artwork
+    roi = image_bgr[y:y+h, x:x+w]
+    if roi.size == 0:
+        return image_bgr
+        
+    # Create a mask for the dark text
+    gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)
+    
+    # Threshold: Pixels darker than 120 become white (the mask), others become black
+    _, mask = cv2.threshold(gray, 120, 255, cv2.THRESH_BINARY_INV)
+    
+    # Dilate the mask slightly to ensure the edges of the letters are fully covered
+    kernel = np.ones((3, 3), np.uint8)
+    mask = cv2.dilate(mask, kernel, iterations=1)
+    
+    # Inpaint the background to erase the text
+    inpainted_roi = cv2.inpaint(roi, mask, inpaintRadius=3, flags=cv2.INPAINT_TELEA)
+    
+    # Paste the cleaned region back into the main image
+    image_bgr[y:y+h, x:x+w] = inpainted_roi
+    
+    return image_bgr
+
+def main():
+    parser = argparse.ArgumentParser(description="Clean manga text using bubbles.json")
+    parser.add_argument("-i", "--image", required=True, help="Path to the original manga image")
+    parser.add_argument("-j", "--json", required=True, help="Path to the bubbles.json file")
+    parser.add_argument("-o", "--output", help="Path to save the cleaned image (optional)")
+    
+    args = parser.parse_args()
+    
+    if not os.path.exists(args.image):
+        print(f"❌ Error: Image file not found at {args.image}")
+        return
+        
+    if not os.path.exists(args.json):
+        print(f"❌ Error: JSON file not found at {args.json}")
+        return
+
+    # Load the image
+    print(f"📂 Loading image: {args.image}")
+    image = cv2.imread(args.image)
+    
+    # Load the JSON data
+    print(f"📂 Loading JSON: {args.json}")
+    with open(args.json, "r", encoding="utf-8") as f:
+        bubbles_data = json.load(f)
+        
+    # Process each box
+    print("🧹 Cleaning text from bounding boxes...")
+    white_count = 0
+    complex_count = 0
+    
+    for box_id, data in bubbles_data.items():
+        if "box" in data:
+            # Default to "complex" if the flag is missing for backward compatibility
+            bg_type = data.get("background_type", "complex")
+            
+            if bg_type == "white":
+                white_count += 1
+            else:
+                complex_count += 1
+                
+            image = clean_text_from_box(image, data["box"], bg_type)
+            
+    print(f"   ✓ Cleaned {white_count} white boxes (fast fill)")
+    print(f"   ✓ Cleaned {complex_count} complex boxes (inpainting)")
+            
+    # Determine output path
+    if args.output:
+        out_path = args.output
+    else:
+        base_name, ext = os.path.splitext(args.image)
+        out_path = f"{base_name}_cleaned{ext}"
+        
+    # Save the result
+    cv2.imwrite(out_path, image)
+    print(f"✅ Cleaned image saved successfully to: {out_path}")
+
+if __name__ == "__main__":
+    main()

manga-translator.py

@@ -1,11 +1,12 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
-
+import cv2
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+import textwrap
 import os
 import re
 import json
-import cv2
-import numpy as np
 import warnings
 from typing import List, Tuple, Dict, Any, Optional
 
@@ -519,6 +520,36 @@ def build_region_flags(raw_text, corrected_text, region_type, conf):
 # ============================================================
 # HELPERS
 # ============================================================
+def inpaint_text_only(image_bgr, box_xywh):
+    """
+    Erases the original text inside the bounding box using inpainting,
+    leaving the background intact without drawing new text.
+    """
+    x, y = int(box_xywh["x"]), int(box_xywh["y"])
+    w, h = int(box_xywh["w"]), int(box_xywh["h"])
+    
+    # 1. Extract the Region of Interest (ROI)
+    roi = image_bgr[y:y+h, x:x+w]
+    if roi.size == 0:
+        return image_bgr
+        
+    # 2. Create a mask for the dark text
+    gray = cv2.cvtColor(roi, cv2.COLOR_BGR2GRAY)
+    # Threshold: Pixels darker than 100 become white (the mask), others become black
+    _, mask = cv2.threshold(gray, 100, 255, cv2.THRESH_BINARY_INV)
+    
+    # Dilate the mask slightly to ensure the edges of the letters are fully covered
+    kernel = np.ones((3,3), np.uint8)
+    mask = cv2.dilate(mask, kernel, iterations=1)
+    
+    # 3. Inpaint the background to erase the text
+    inpainted_roi = cv2.inpaint(roi, mask, inpaintRadius=3, flags=cv2.INPAINT_TELEA)
+    
+    # Put the erased background back into the main image
+    image_bgr[y:y+h, x:x+w] = inpainted_roi
+    
+    return image_bgr
+
 def split_boxes_by_horizontal_gap(out_boxes, out_indices, out_quads, out_lines, ocr, gap_multiplier=1.5):
     """
     Splits a single bounding box into multiple boxes if there is a large horizontal 
@@ -2749,6 +2780,14 @@ def process_manga_page(image_path: str,
         # Bubble groups (lines as rendered in the bubble)
         bubble_groups = build_text_from_layout(indices, ocr)
 
+        # ── Step 12.5: Detect Background Complexity ───────────
+        # Analyze the pixels to see if it's a plain white bubble or complex artwork
+        feats = contour_features_for_box(image_bgr, adjusted_box_xyxy)
+        if feats["whiteness_ratio"] > 0.75 and feats["edge_density"] < 0.10:
+            bg_type = "white"
+        else:
+            bg_type = "complex"
+
         # ── Step 13: Translate ────────────────────────────────
         translated        = ""
         translation_input = corrected_text
@@ -2779,6 +2818,7 @@ def process_manga_page(image_path: str,
         results[str(bid)] = {
             "order":             order_idx,
             "region_type":       region_type,
+            "background_type":   bg_type,              # <--- NEW FLAG ADDED HERE
             "confidence":        round(conf, 4),
             "ocr_source":        ocr_source,
             "raw_ocr":           raw_text,
@@ -2787,7 +2827,7 @@ def process_manga_page(image_path: str,
             "translated":        translated,
             "flags":             flags,
             "bubble_groups":     bubble_groups,
-            "box":               xyxy_to_xywh(adjusted_box_xyxy), # <--- Uses the adjusted box
+            "box":               xyxy_to_xywh(adjusted_box_xyxy),
             "lines":             bubble_groups,
         }
 
@@ -2801,6 +2841,7 @@ def process_manga_page(image_path: str,
         _write_txt_output(results, output_txt)
 
     return results
+
 # ============================================================
 # OUTPUT WRITERS
 # ============================================================