llm_ticket3/utils/ocr_preprocessor.py

"""
!!! FICHIER OBSOLÈTE !!!
Ce fichier est maintenu uniquement pour la compatibilité.
Veuillez utiliser les modules dans utils/ocr_brut/ à la place.
"""

# ocr_preprocessor.py

import os
from PIL import Image, ImageEnhance
import cv2
import numpy as np
import time

# === 🎛️ PROFILS DE TRAITEMENT D'IMAGE OCR ===
PREPROCESSING_PROFILES = {
    "default": {
        "resize_min_dim": 1000,        # 📏 Si largeur/hauteur < 1000px, image agrandie proportionnellement
        "enhance_contrast": True,      # 🔆 Active l'amélioration du contraste
        "contrast_factor": 1.2,        # >1 = plus contrasté, typique : 1.2 à 1.8
        "enhance_sharpness": False,    # 💥 Active la netteté
        "sharpness_factor": 1.0,       # >1 = plus net, typique : 1.2 à 2.0
        "apply_denoising": False,      # 🚿 Réduction de bruit
        "denoise_strength": {
            "h": 0,                    # 0 à 15 : intensité du lissage luminance
            "hColor": 0,               # 0 à 15 : lissage chroma
            "templateWindowSize": 7,   # Taille du patch à comparer (typiquement 7)
            "searchWindowSize": 21     # Zone autour du patch pour recherche (typiquement 21)
        },
        "invert_colors": False,        # ↕️ Inversion si texte clair sur fond sombre
        "apply_clahe": False,          # 📈 Égalisation du contraste local (utile en cas de zones très sombres/claires)
        "save_debug_output": True,
        "debug_output_dir": "debug_ocr_preprocessed"
    },
    "aggressive": {
        "resize_min_dim": 1400,
        "enhance_contrast": True,
        "contrast_factor": 1.8,
        "enhance_sharpness": True,
        "sharpness_factor": 1.5,
        "apply_denoising": True,
        "denoise_strength": {
            "h": 10,
            "hColor": 10,
            "templateWindowSize": 7,
            "searchWindowSize": 21
        },
        "invert_colors": False,
        "apply_clahe": False,
        "save_debug_output": True,
        "debug_output_dir": "debug_ocr_preprocessed"
    },
    "document": {
        "resize_min_dim": 1100,
        "enhance_contrast": True,
        "contrast_factor": 1.2,
        "enhance_sharpness": False,
        "sharpness_factor": 1.0,
        "apply_denoising": False,
        "denoise_strength": {"h": 0, "hColor": 0, "templateWindowSize": 7, "searchWindowSize": 21},
        "invert_colors": False,
        "apply_clahe": False,
        "save_debug_output": True,
        "debug_output_dir": "debug_ocr_preprocessed"
    },
    "clahe_high": {
        "resize_min_dim": 1200,
        "enhance_contrast": True,
        "contrast_factor": 1.4,
        "enhance_sharpness": True,
        "sharpness_factor": 1.3,
        "apply_denoising": True,
        "denoise_strength": {
            "h": 7,
            "hColor": 7,
            "templateWindowSize": 7,     # Taille du patch local utilisé
            "searchWindowSize": 21       # Zone de recherche du filtre
        },
        "invert_colors": False,
        "apply_clahe": True,
        "save_debug_output": True,
        "debug_output_dir": "debug_ocr_preprocessed"
    },
    "invert_light": {
        "resize_min_dim": 1200,
        "enhance_contrast": True,
        "contrast_factor": 1.3,
        "enhance_sharpness": True,
        "sharpness_factor": 1.4,
        "apply_denoising": False,
        "invert_colors": True,
        "apply_clahe": False,
        "save_debug_output": True,
        "debug_output_dir": "debug_ocr_preprocessed"
    }
}

def preprocess_image(image_path: str, **settings) -> Image.Image:
    img = Image.open(image_path).convert("RGB")
    base_name = os.path.basename(image_path)
    
    # Gestion des dossiers de debug
    debug_dir = settings.get("debug_output_dir", "debug_ocr_preprocessed")
    profile_name = settings.get("profile_name", "default")  # Ajout du nom du profil
    debug_profile_dir = os.path.join(debug_dir, profile_name)
    os.makedirs(debug_profile_dir, exist_ok=True)

    # Redimensionnement
    if settings.get("resize_min_dim", 0) > 0:
        width, height = img.size
        min_dim = min(width, height)
        if min_dim < settings["resize_min_dim"]:
            scale = settings["resize_min_dim"] / min_dim
            new_size = (int(width * scale), int(height * scale))
            img = img.resize(new_size, Image.Resampling.BICUBIC)

    # Contraste
    if settings.get("enhance_contrast", False):
        enhancer = ImageEnhance.Contrast(img)
        img = enhancer.enhance(settings.get("contrast_factor", 1.5))

    # Netteté
    if settings.get("enhance_sharpness", False):
        enhancer = ImageEnhance.Sharpness(img)
        img = enhancer.enhance(settings.get("sharpness_factor", 1.5))

    # Convert to OpenCV image
    img_cv = cv2.cvtColor(np.array(img), cv2.COLOR_RGB2BGR)

    # Débruitage
    if settings.get("apply_denoising", False):
        strength = settings.get("denoise_strength", {})
        img_cv = cv2.fastNlMeansDenoisingColored(
            img_cv,
            None,
            h=strength.get("h", 10),
            hColor=strength.get("hColor", 10),
            templateWindowSize=strength.get("templateWindowSize", 7),
            searchWindowSize=strength.get("searchWindowSize", 21)
        )

    # CLAHE
    if settings.get("apply_clahe", False):
        lab = cv2.cvtColor(img_cv, cv2.COLOR_BGR2LAB)
        l, a, b = cv2.split(lab)
        clahe = cv2.createCLAHE(clipLimit=2.0, tileGridSize=(8, 8))
        cl = clahe.apply(l)
        img_cv = cv2.merge((cl, a, b))
        img_cv = cv2.cvtColor(img_cv, cv2.COLOR_LAB2BGR)

    # Inversion
    if settings.get("invert_colors", False):
        img_cv = cv2.bitwise_not(img_cv)

    # Sauvegarde image prétraitée (debug)
    if settings.get("save_debug_output", False):
        debug_path = os.path.join(debug_profile_dir, f"optimized_{base_name}")
        cv2.imwrite(debug_path, img_cv)

    return Image.fromarray(cv2.cvtColor(img_cv, cv2.COLOR_BGR2RGB))

def preprocess_image_with_profile(image_path: str, profile_name="default") -> Image.Image:
    settings = PREPROCESSING_PROFILES[profile_name].copy()  # On fait une copie pour ne pas modifier l'original
    settings["profile_name"] = profile_name  # On ajoute le nom du profil aux paramètres
    return preprocess_image(image_path, **settings)