Update app.py

app.py

@@ -1,40 +1,89 @@
 from fastapi import FastAPI, UploadFile, File
-import tensorflow as tf
+
 import numpy as np
+import cv2
+import tensorflow as tf
 from PIL import Image
 import io
-import uvicorn
+
 
 app = FastAPI()
 
-# Load model Keras
-model = tf.keras.models.load_model("lontara_model_finetuning.keras")
-
-# Label kelas sesuai model
-labels = [
-    "Tu", "He", "We", "No", "Mu", "Bu", "Ji", "Jo", "I", "Nro", "Cu", "Na", "Bo", "Yi", "Se", "Nyi",
-    "So", "Wa", "Ko", "Ge", "E", "Yo", "Ngu", "Ra", "Wo", "Ta", "Pe", "Nra", "Da", "Ci", "Lo", "Nci",
-    "U", "Ro", "Mo", "Nre", "Du", "Be", "Mpu", "Hu", "Ne", "Nyo", "Ncu", "Su", "Ju", "Gu", "Nu", "Di",
-    "Nri", "Gi", "Co", "Nca", "Ri", "Si", "Ja", "Bi", "Ke", "Wu", "Nki", "Te", "Go", "Ya", "Nku", "Pu",
-    "Nka", "Ba", "Mpe", "A", "Nya", "Me", "Nge", "Mpa", "Ma", "Mpi", "O", "Mi", "Re", "Po", "Ti", "Je",
-    "Nco", "Pa", "Ho", "Nko", "Ce", "Li", "Nke", "Ru", "Ca", "Ke_", "Do", "Ga", "Mpo", "Nye", "Nru", "Nga",
-    "Lu", "Pi", "Ku", "Ni", "Nce", "Le", "Ngo", "De", "Ki", "Wi", "Hi", "Ye", "Ngi", "Ka", "Nyu", "La",
-    "Ha", "Sa"
-]
-
-@app.get("/")
 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
-    image = Image.open(io.BytesIO(await file.read())).convert("L").resize((128, 128))
-    image = np.array(image) / 255.0
-    image = image.reshape(1, 128, 128, 1)
+    # 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
 
-    # Prediksi
-    prediction = model.predict(image)
-    label = labels[np.argmax(prediction)]
+        prediction = model.predict(char_norm)
+        predicted_label = labels[np.argmax(prediction)]
+        predictions.append(predicted_label)
 
-    return {"prediction": label}
+    return {"predictions": predictions}