utils.py

import re
import numpy as np
import cv2
from PIL import Image
import random
import torch
import torchvision.transforms as T
from torchvision.transforms.functional import InterpolationMode
from difflib import SequenceMatcher
from nltk.metrics.distance import edit_distance
import nltk

# Ensure NLTK data is downloaded
try:
    nltk.data.find('corpora/words.zip')
except LookupError:
    nltk.download('words')
try:
    nltk.data.find('tokenizers/punkt')
except LookupError:
    nltk.download('punkt')

from nltk.corpus import words

def set_seed(seed=42):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    # torch.cuda.manual_seed_all(seed) # Uncomment if using GPU
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = False

def build_transform(input_size=448):
    mean = (0.485, 0.456, 0.406)
    std = (0.229, 0.224, 0.225)
    return T.Compose([
        T.Lambda(lambda img: img.convert('RGB')),
        T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
        T.ToTensor(),
        T.Normalize(mean=mean, std=std)
    ])

def get_roi(image_path_or_obj, *roi):
    """
    Extracts ROI from an image path or PIL Image object.
    """
    if isinstance(image_path_or_obj, str):
        image = Image.open(image_path_or_obj).convert('RGB')
    else:
        image = image_path_or_obj.convert('RGB')
        
    width, height = image.size

    roi_x_start = int(width * roi[0])
    roi_y_start = int(height * roi[1])
    roi_x_end = int(width * roi[2])
    roi_y_end = int(height * roi[3])

    cropped_image = image.crop((roi_x_start, roi_y_start, roi_x_end, roi_y_end))
    return cropped_image

def clean_text(text):
    return re.sub(r'[^a-zA-Z0-9]', '', text).strip().lower()

def are_strings_similar(str1, str2, max_distance=3, max_length_diff=2):
    if str1 == str2:
        return True
    if abs(len(str1) - len(str2)) > max_length_diff:
        return False
    edit_distance_value = edit_distance(str1, str2)
    return edit_distance_value <= max_distance

def blur_image(image, strength):
    image_np = np.array(image)
    blur_strength = int(strength * 50)
    blur_strength = max(1, blur_strength | 1)
    blurred_image = cv2.GaussianBlur(image_np, (blur_strength, blur_strength), 0)
    blurred_pil_image = Image.fromarray(blurred_image)
    return blurred_pil_image

def is_blank(text, limit=15):
    return len(text) < limit

def string_similarity(a, b):
    return SequenceMatcher(None, a.lower(), b.lower()).ratio()

def find_similar_substring(text, keyword, threshold=0.9):
    text = text.lower()
    keyword = keyword.lower()

    if keyword in text:
        return True

    keyword_length = len(keyword.split())
    words_list = text.split()

    for i in range(len(words_list) - keyword_length + 1):
        phrase = ' '.join(words_list[i:i + keyword_length])
        similarity = string_similarity(phrase, keyword)
        if similarity >= threshold:
            return True

    return False

def destroy_text_roi(image, *roi_params):
    image_np = np.array(image)

    h, w, _ = image_np.shape
    x1 = int(roi_params[0] * w)
    y1 = int(roi_params[1] * h)
    x2 = int(roi_params[2] * w)
    y2 = int(roi_params[3] * h)

    roi = image_np[y1:y2, x1:x2]

    blurred_roi = cv2.GaussianBlur(roi, (75, 75), 0)
    noise = np.random.randint(0, 50, (blurred_roi.shape[0], blurred_roi.shape[1], 3), dtype=np.uint8)
    noisy_blurred_roi = cv2.add(blurred_roi, noise)
    image_np[y1:y2, x1:x2] = noisy_blurred_roi
    return Image.fromarray(image_np)

def is_english(text):
    allowed_pattern = re.compile(
        r'^[a-zA-Z०-९\u0930\s\.,!?\-;:"\'()]*$'
    )
    return bool(allowed_pattern.match(text))

def is_valid_english(text):
    english_words = set(words.words())
    cleaned_words = ''.join(c.lower() if c.isalnum() else ' ' for c in text).split()
    return all(word.lower() in english_words for word in cleaned_words)