malayalam-ocr / app.py
afueth
Updated files and ico
443c940
Raw History Blame Contribute Delete
4.59 kB
from __future__ import annotations
import io
import streamlit as st
import fitz # PyMuPDF
from PIL import Image
from ocr.ocr_engine import configure_tesseract, ocr_pil_image
from ocr.pdf_utils import pdf_has_native_text, extract_native_text, render_page_to_pil
from ocr.image_utils import preprocess_image
st.set_page_config(page_title="Malayalam OCR", page_icon="favicon.ico", layout="wide")
st.title("Malayalam OCR")
configure_tesseract()
with st.sidebar:
st.header("OCR Settings")
lang = st.selectbox("Language", ["mal", "mal+eng"], index=0)
psm = st.selectbox("PSM (layout mode)", [3, 4, 6, 11], index=2)
zoom = st.slider("PDF render zoom (OCR quality)", 1.0, 3.5, 2.0, 0.1)
st.divider()
st.header("Preprocessing")
grayscale = st.checkbox("Grayscale", value=True)
contrast = st.slider("Contrast", 1.0, 3.0, 2.0, 0.1)
use_threshold = st.checkbox("Threshold", value=True)
threshold = st.slider("Threshold value", 0, 255, 160, 5) if use_threshold else None
invert = st.checkbox("Invert colors", value=False)
st.divider()
st.header("PDF Behavior")
force_ocr = st.checkbox("Force OCR (use for scanned PDFs)", value=False)
max_pages = st.slider("Max pages to process", 1, 200, 20)
def do_ocr_on_pil(img: Image.Image) -> str:
img2 = preprocess_image(
img,
grayscale=grayscale,
contrast=contrast,
threshold=threshold,
invert=invert,
)
return ocr_pil_image(img2, lang=lang, psm=int(psm), oem=1)
@st.cache_data(show_spinner=False)
def process_image_bytes(img_bytes: bytes, settings_key: str) -> str:
img = Image.open(io.BytesIO(img_bytes))
return do_ocr_on_pil(img)
@st.cache_data(show_spinner=False)
def process_pdf_bytes(pdf_bytes: bytes, settings_key: str) -> str:
doc = fitz.open(stream=pdf_bytes, filetype="pdf")
out = []
total = min(doc.page_count, max_pages)
for i in range(total):
page = doc.load_page(i)
if (not force_ocr) and pdf_has_native_text(page, min_chars=50):
txt = extract_native_text(page)
out.append(f"\n\n--- Page {i+1} (native text) ---\n{txt}")
else:
pil_img = render_page_to_pil(page, zoom=zoom)
txt = do_ocr_on_pil(pil_img)
out.append(f"\n\n--- Page {i+1} (OCR) ---\n{txt}")
return "\n".join(out).strip()
uploaded = st.file_uploader("Upload PDF or Image", type=["pdf", "png", "jpg", "jpeg", "webp"])
if uploaded:
settings_key = f"{lang}|{psm}|{zoom}|{grayscale}|{contrast}|{threshold}|{invert}|{force_ocr}|{max_pages}"
file_bytes = uploaded.getvalue()
name = uploaded.name.lower()
col1, col2 = st.columns([1, 1])
with col1:
st.subheader("Input Preview")
if name.endswith(".pdf"):
st.info("PDF uploaded. Previewing first page as image.")
doc = fitz.open(stream=file_bytes, filetype="pdf")
page0 = doc.load_page(0)
img0 = render_page_to_pil(page0, zoom=1.5)
st.image(img0, use_container_width=True)
else:
img = Image.open(io.BytesIO(file_bytes))
st.image(img, use_container_width=True)
with col2:
st.subheader("Extracted Text")
with st.spinner("Extracting text..."):
if name.endswith(".pdf"):
text = process_pdf_bytes(file_bytes, settings_key)
else:
text = process_image_bytes(file_bytes, settings_key)
st.text_area("Copy text from here", value=text, height=420, key="result_text_area")
# ✅ Copy button (clipboard) — works in browser
copy_disabled = not bool(text.strip())
copy_label = "Copy" if not copy_disabled else "Copy (no text)"
clicked = st.button(copy_label, use_container_width=True, disabled=copy_disabled)
if clicked and not copy_disabled:
st.components.v1.html(
f"""
<script>
navigator.clipboard.writeText({text!r});
</script>
""",
height=0,
)
st.success("Copied to clipboard!")
st.download_button(
"Download as .txt",
data=text.encode("utf-8"),
file_name=uploaded.name.rsplit(".", 1)[0] + "_ocr.txt",
mime="text/plain",
use_container_width=True,
)
st.caption("Tip: If results are messy, try PSM=4 or PSM=3, increase Zoom to 2.5–3.0, or adjust Threshold.")
else:
st.write("Upload a PDF or image to extract Malayalam printed text.")