app.py

import streamlit as st
from PIL import Image
import pytesseract
import io
import fitz  # PyMuPDF
import cv2
import numpy as np
import requests
from transformers import pipeline
from difflib import SequenceMatcher
import folium
from streamlit_folium import st_folium
import wikipedia

# Load summarization and NER pipeline
summarizer = pipeline("summarization", model="facebook/bart-large-cnn")
ner_pipeline = pipeline("ner", aggregation_strategy="simple")

# Streamlit App
st.set_page_config(page_title="AI Historical Document Decipher", layout="wide")
st.title("📜 AI-powered Historical Document Deciphering App")

st.sidebar.header("Upload Document")
uploaded_file = st.sidebar.file_uploader("Upload Image or PDF", type=["jpg", "jpeg", "png", "pdf"])

# Function to convert PDF to image
def pdf_to_images(pdf_bytes):
    doc = fitz.open(stream=pdf_bytes, filetype="pdf")
    images = []
    for page in doc:
        pix = page.get_pixmap()
        img = Image.open(io.BytesIO(pix.tobytes()))
        images.append(img)
    return images

# Function to enhance image
def enhance_image(image):
    img = np.array(image.convert('RGB'))
    gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

    # Denoise
    denoised = cv2.fastNlMeansDenoising(gray, h=30)

    # Sharpening
    kernel = np.array([[0, -1, 0],
                       [-1, 5,-1],
                       [0, -1, 0]])
    sharpened = cv2.filter2D(denoised, -1, kernel)

    # Thresholding (binarization)
    _, binary = cv2.threshold(sharpened, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)

    # Optional: Resize (sometimes helps OCR)
    scale_percent = 150  # percent of original size
    width = int(binary.shape[1] * scale_percent / 100)
    height = int(binary.shape[0] * scale_percent / 100)
    resized = cv2.resize(binary, (width, height), interpolation=cv2.INTER_CUBIC)

    return resized

# Function to perform OCR
def perform_ocr(image):
    custom_oem_psm_config = r'--oem 3 --psm 6 -c preserve_interword_spaces=1'
    text = pytesseract.image_to_string(image, config=custom_oem_psm_config)
    return text

# Function to extract named entities
def extract_entities(text):
    entities = ner_pipeline(text)
    extracted = {}
    for ent in entities:
        label = ent['entity_group']
        extracted.setdefault(label, set()).add(ent['word'])
    return extracted

def get_historical_context(entities):
    context = {}
    for label, values in entities.items():
        for item in values:
            try:
                summary = wikipedia.summary(item, sentences=2)
                context[item] = summary
            except wikipedia.exceptions.DisambiguationError as e:
                context[item] = f"Multiple entries found for '{item}': {e.options[:3]}"
            except wikipedia.exceptions.PageError:
                context[item] = f"No historical info found for '{item}'."
            except Exception as e:
                context[item] = f"Error retrieving info: {e}"
    return context

# Function to correct OCR errors (suggestions)
def suggest_corrections(original_text):
    words = original_text.split()
    suggestions = {}
    for word in words:
        if len(word) > 4 and not word.isnumeric():
            close_matches = [w for w in ["document", "historical", "archive", "event", "location"] if SequenceMatcher(None, word.lower(), w).ratio() > 0.75]
            if close_matches:
                suggestions[word] = close_matches[0]
    return suggestions

# Function to generate map
def generate_map(entities):
    m = folium.Map(location=[20, 0], zoom_start=2)
    if "LOC" in entities:
        for location in entities["LOC"]:
            # Dummy coordinates for demonstration
            folium.Marker(
                location=[51.5074, -0.1278],  # Example: London
                popup=f"Location: {location}",
                tooltip=location
            ).add_to(m)
    return m

if uploaded_file:
    file_type = uploaded_file.type

    # Display and process the uploaded document
    if file_type == "application/pdf":
        images = pdf_to_images(uploaded_file.read())
    else:
        images = [Image.open(uploaded_file)]

    for image in images:
        st.image(image, caption="Uploaded Document", use_container_width=True)

        # Enhance image
        enhanced = enhance_image(image)
        st.image(enhanced, caption="Enhanced Image", use_container_width=True, channels="GRAY")

        # Perform OCR
        ocr_text = perform_ocr(enhanced)
        st.subheader("Extracted Text (OCR)")
        st.text_area("Text", ocr_text, height=200)

        # Suggest corrections
        corrections = suggest_corrections(ocr_text)
        if corrections:
            st.subheader("AI Suggestions for Possible Corrections")
            for original, suggestion in corrections.items():
                st.markdown(f"**{original}** ➔ *{suggestion}*")

        # Summarize text
        if len(ocr_text.strip()) > 50:
            summary = summarizer(ocr_text, max_length=60, min_length=20, do_sample=False)[0]['summary_text']
            st.subheader("Summary")
            st.write(summary)

        # Extract entities
        entities = extract_entities(ocr_text)
        st.subheader("Key Information")
        for label, items in entities.items():
            st.markdown(f"**{label}**: {', '.join(items)}")

        # Provide historical context
        context = get_historical_context(entities)
        if context:
            st.subheader("Historical Context & Insights")
            for item, info in context.items():
                st.markdown(f"- **{item}**: {info}")

        # Visualize map
        st.subheader("Locations Mentioned")
        map_ = generate_map(entities)
        st_folium(map_, width=700)

        st.markdown("---")

else:
    st.info("Upload an image or PDF of a historical document to begin.")

st.sidebar.markdown("---")
st.sidebar.markdown("Developed by **Cherilyn Marie Deocampo**")