app.py

from fastapi import FastAPI, UploadFile, File

import numpy as np
import cv2
import tensorflow as tf
from PIL import Image
import io


app = FastAPI()

# Load model Keras
model = tf.keras.models.load_model("lontara_model_finetuning.keras")

def home():
    return {"message": "Aksara Lontara API is running"}

def preprocess_image(image: np.ndarray):
    """ Melakukan segmentasi karakter menggunakan OpenCV """

    # **1️⃣ Edge Detection (Canny)**
    edges = cv2.Canny(image, 50, 150)

    # **2️⃣ Morphological Cleaning**
    kernel = np.ones((3, 3), np.uint8)
    edges_cleaned = cv2.morphologyEx(edges, cv2.MORPH_CLOSE, kernel, iterations=2)

    # **3️⃣ Connected Component Analysis (CCA)**
    num_labels, labels, stats, _ = cv2.connectedComponentsWithStats(edges_cleaned, connectivity=8)

    # **4️⃣ Filter huruf berdasarkan area**
    min_area = 500
    bounding_boxes = []

    for i in range(1, num_labels):  # Skip background
        x, y, w, h, area = stats[i]
        if area > min_area:
            bounding_boxes.append((x, y, w, h))

    # **5️⃣ Urutkan huruf berdasarkan posisi X**
    bounding_boxes.sort(key=lambda b: b[0])

    # **6️⃣ Gabungkan Bounding Box yang Berdekatan**
    merged_boxes = []
    merge_threshold = 20

    for i in range(len(bounding_boxes)):
        x, y, w, h = bounding_boxes[i]

        if merged_boxes and (x - (merged_boxes[-1][0] + merged_boxes[-1][2])) < merge_threshold:
            x_prev, y_prev, w_prev, h_prev = merged_boxes.pop()
            x_new = min(x_prev, x)
            y_new = min(y_prev, y)
            w_new = max(x_prev + w_prev, x + w) - x_new
            h_new = max(y_prev + h_prev, y + h) - y_new
            merged_boxes.append((x_new, y_new, w_new, h_new))
        else:
            merged_boxes.append((x, y, w, h))

    # **7️⃣ Potong dan proses karakter**
    segmented_chars = []
    for (x, y, w, h) in merged_boxes:
        char_segment = image[y:y+h, x:x+w]
        char_segment = cv2.resize(char_segment, (128, 128), interpolation=cv2.INTER_AREA)
        segmented_chars.append(char_segment)

    return segmented_chars

@app.post("/predict")
async def predict(file: UploadFile = File(...)):
    # Baca gambar dari file upload
    image = Image.open(io.BytesIO(await file.read())).convert("L")
    image = np.array(image)

    # **Segmentasi huruf**
    segmented_chars = preprocess_image(image)

    # Jika tidak ada huruf terdeteksi
    if not segmented_chars:
        return {"prediction": "No characters detected"}

    # **Prediksi untuk setiap karakter**
    predictions = []
    for char in segmented_chars:
        char_norm = np.array(char) / 255.0  # Normalisasi
        char_norm = char_norm.reshape(1, 128, 128, 1)  # Reshape untuk model

        prediction = model.predict(char_norm)
        predicted_label = labels[np.argmax(prediction)]
        predictions.append(predicted_label)

    return {"predictions": predictions}