app/main.py

from fastapi import FastAPI, UploadFile, File, Form, HTTPException
from fastapi.responses import JSONResponse, FileResponse
from pydantic import BaseModel
import io
from PIL import Image
import json
import tempfile
import base64
from typing import List, Optional, Any

from kraken import binarization
from kraken import pageseg
from kraken import rpred
from kraken.lib import models
from kraken import blla
from kraken import serialization
from kraken.lib.exceptions import KrakenInvalidModelException

import numpy as np

app = FastAPI()

class RawResponse(BaseModel):
    result: Any

def serialize_line(line):
    # Create a dictionary with all available attributes
    line_dict = vars(line)
    
    # If 'bbox' is not available but 'polygon' is, calculate bbox from polygon
    if 'bbox' not in line_dict and 'polygon' in line_dict and line_dict['polygon'] is not None:
        x_coords, y_coords = zip(*line_dict['polygon'])
        bbox = [min(x_coords), min(y_coords), max(x_coords), max(y_coords)]
        line_dict['bbox'] = bbox
    
    # Convert numpy arrays to lists for JSON serialization
    for key, value in line_dict.items():
        if isinstance(value, np.ndarray):
            line_dict[key] = value.tolist()
    
    return line_dict

def serialize_region(region):
    # Create a dictionary with known attributes
    region_dict = {
        "id": getattr(region, 'id', None),
        "boundary": getattr(region, 'boundary', None),
        "tags": getattr(region, 'tags', None),
    }
    
    # Convert numpy arrays to lists for JSON serialization
    for key, value in region_dict.items():
        if isinstance(value, np.ndarray):
            region_dict[key] = value.tolist()
    
    return region_dict

@app.post("/detect_lines", response_model=RawResponse)
async def detect_lines(file: UploadFile = File(...)):
    content = await file.read()
    image = Image.open(io.BytesIO(content))
    
    # Perform baseline and layout analysis (BLLA) segmentation with default model
    baseline_seg = blla.segment(image)
    
    serialized_seg = {
        "lines": [serialize_line(line) for line in baseline_seg.lines],
        "regions": [serialize_region(region) for region in baseline_seg.regions],
        "type": baseline_seg.type,
        "text_direction": baseline_seg.text_direction,
        "script_detection": baseline_seg.script_detection,
    }
    
    return RawResponse(result=serialized_seg)

@app.post("/ocr", response_model=RawResponse)
async def perform_ocr(
    file: UploadFile = File(...),
    model_name: str = Form("catmus-medieval.mlmodel"),
    binarize: bool = Form(False)
):
    content = await file.read()
    image = Image.open(io.BytesIO(content))
    
    # Always binarize the image before segmentation
    bw_img = binarization.nlbin(image)
    
    try:
        model = models.load_any(model_name)
    except KrakenInvalidModelException:
        raise HTTPException(status_code=400, detail=f"Model '{model_name}' not found or invalid")
    
    baseline_seg = pageseg.segment(bw_img)
    
    # Use the original image for OCR if binarize is False, otherwise use the binarized image
    ocr_image = bw_img if binarize else image
    
    result = list(rpred.rpred(model, ocr_image, baseline_seg))
    
    serialized_result = [
        {
            "bbox": record.bbox,
            # "confidence": record.confidence,
            "text": record.prediction,
            "cuts": record.cuts,
            # "line_id": record.line_id,
        }
        for record in result
    ]
    
    return RawResponse(result=serialized_result)

@app.post("/segment", response_model=RawResponse)
async def segment_image(
    file: UploadFile = File(...),
    baseline: bool = Form(True)
):
    content = await file.read()
    image = Image.open(io.BytesIO(content))
    
    bw_img = binarization.nlbin(image)
    
    if baseline:
        segmentation = pageseg.segment(bw_img)
    else:
        segmentation = pageseg.segment(bw_img, text_direction='horizontal-lr')
    
    serialized_seg = {
        "lines": [serialize_line(line) for line in segmentation.lines],
        "regions": [vars(region) for region in segmentation.regions],
        "type": segmentation.type,
        "text_direction": segmentation.text_direction,
        "script_detection": segmentation.script_detection,
    }
    
    return RawResponse(result=serialized_seg)

@app.post("/binarize")
async def binarize_image(file: UploadFile = File(...)):
    content = await file.read()
    image = Image.open(io.BytesIO(content))
    
    bw_img = binarization.nlbin(image)
    
    with tempfile.NamedTemporaryFile(suffix=".png", delete=False) as temp_file:
        bw_img.save(temp_file.name)
        return FileResponse(temp_file.name, media_type="image/png", filename="binarized.png")

@app.post("/process_all", response_model=RawResponse)
async def process_all(
    file: UploadFile = File(...),
    model_name: str = Form("catmus-medieval")
):
    content = await file.read()
    image = Image.open(io.BytesIO(content))
    
    # Step 1: Binarization
    bw_img = binarization.nlbin(image)
    
    # Convert binarized image to base64 for JSON response
    buffered = io.BytesIO()
    bw_img.save(buffered, format="PNG")
    binarized_base64 = base64.b64encode(buffered.getvalue()).decode()
    
    # Step 2: Segmentation
    segmentation = pageseg.segment(bw_img)
    serialized_seg = {
        "lines": [serialize_line(line) for line in segmentation.lines],
        "regions": [vars(region) for region in segmentation.regions],
        "type": segmentation.type,
        "text_direction": segmentation.text_direction,
        "script_detection": segmentation.script_detection,
    }
    
    # Step 3: OCR
    try:
        model = models.load_any(model_name)
    except KrakenInvalidModelException:
        raise HTTPException(status_code=400, detail=f"Model '{model_name}' not found or invalid")
    
    result = list(rpred.rpred(model, bw_img, segmentation))
    print(result)
    serialized_result = [
        {
            "bbox": record.bbox,
            # "confidence": record.confidence,
            "text": record.prediction,
            "cuts": record.cuts,
            # "line_id": record.line_id,
        }
        for record in result
    ]
    
    return RawResponse(result={
        "binarized_image": binarized_base64,
        "segmentation": serialized_seg,
        "ocr_result": serialized_result
    })

@app.get("/")
async def root():
    return {
        "message": "Welcome to the Complete Kraken Python API",
        "available_endpoints": ["/detect_lines", "/ocr", "/segment", "/binarize", "/process_all"]
    }

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860, workers=1)