Refactor preprocessing pipeline to streamline image processing and remove unused background removal logic

app.py

@@ -6,7 +6,6 @@ import io
 from flask import Flask, request, render_template
 from PIL import Image
 from skimage.feature import local_binary_pattern
-from rembg import remove
 
 app = Flask(__name__)
 
@@ -57,80 +56,16 @@ def extract_lbp_features(gray_img, P=8, R=1, method="uniform"):
     return hist
 
 
-def preprocess_image_like_dataset(pil_img, target_size=100):
+def preprocessing_pipeline(pil_img):
     """
-    Preprocessing gambar dari luar dataset menjadi seperti dataset:
-    - Background putih
-    - Ukuran 100x100
-    - Fokus pada buah (centered dan cropped)
+    Pipeline preprocessing sesuai dengan data training:
+    - Convert ke RGB
+    - Resize ke 100x100
+    - Extract color histogram & LBP features
     """
-    # Step 1: Hapus background menggunakan rembg (AI-powered background removal)
-    img_no_bg = remove(pil_img)
-    img_array = np.array(img_no_bg)
-    
-    # Step 2: Konversi RGBA ke RGB dengan background putih
-    if img_array.shape[-1] == 4:
-        # Ambil channel alpha
-        alpha = img_array[:, :, 3] / 255.0
-        rgb = img_array[:, :, :3]
-        
-        # Gabungkan dengan background putih
-        white_bg = np.ones_like(rgb) * 255
-        img_array = (rgb * alpha[:, :, np.newaxis] + white_bg * (1 - alpha[:, :, np.newaxis])).astype(np.uint8)
-    
-    # Step 3: Crop ke bounding box objek (hilangkan whitespace)
-    # Konversi ke grayscale untuk deteksi objek
-    gray = cv2.cvtColor(img_array, cv2.COLOR_RGB2GRAY)
-    # Threshold untuk menemukan objek (non-putih)
-    _, thresh = cv2.threshold(gray, 250, 255, cv2.THRESH_BINARY_INV)
-    
-    # Temukan contours
-    contours, _ = cv2.findContours(thresh, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
-    
-    if contours:
-        # Gabungkan semua contours untuk mendapat bounding box total
-        all_contours = np.vstack(contours)
-        x, y, w, h = cv2.boundingRect(all_contours)
-        
-        # Tambahkan padding 5% untuk memberikan sedikit ruang
-        padding = int(max(w, h) * 0.05)
-        x = max(0, x - padding)
-        y = max(0, y - padding)
-        w = min(img_array.shape[1] - x, w + 2 * padding)
-        h = min(img_array.shape[0] - y, h + 2 * padding)
-        
-        # Crop gambar
-        img_cropped = img_array[y:y+h, x:x+w]
-    else:
-        img_cropped = img_array
-    
-    # Step 4: Resize dengan mempertahankan aspect ratio dan center pada canvas putih 100x100
-    h, w = img_cropped.shape[:2]
-    
-    # Hitung scaling factor (fit ke target_size dengan margin)
-    scale = (target_size * 0.9) / max(h, w)  # 0.9 untuk margin
-    new_w = int(w * scale)
-    new_h = int(h * scale)
-    
-    # Resize objek
-    img_resized = cv2.resize(img_cropped, (new_w, new_h), interpolation=cv2.INTER_AREA)
-    
-    # Buat canvas putih 100x100
-    canvas = np.ones((target_size, target_size, 3), dtype=np.uint8) * 255
-    
-    # Hitung posisi untuk center objek
-    y_offset = (target_size - new_h) // 2
-    x_offset = (target_size - new_w) // 2
-    
-    # Letakkan objek di tengah canvas
-    canvas[y_offset:y_offset+new_h, x_offset:x_offset+new_w] = img_resized
-    
-    return canvas
-
-
-def preprocessing_pipeline(pil_img):
-    # Preprocessing gambar menjadi 100x100 dengan background putih
-    img = preprocess_image_like_dataset(pil_img, target_size=100)
+    # Convert ke RGB dan resize ke 100x100 (sesuai dataset training)
+    img = np.array(pil_img.convert('RGB'))
+    img = cv2.resize(img, (100, 100), interpolation=cv2.INTER_AREA)
 
     img_float = img.astype(np.float32) / 255.0
     img_uint8 = (img_float * 255).astype(np.uint8)
@@ -167,11 +102,11 @@ def index():
             try:
                 image = Image.open(file.stream)
                 
-                # Preprocessing image untuk prediksi dan tampilan
+                # Preprocessing sesuai dengan data training (hanya resize 100x100)
                 features = preprocessing_pipeline(image)
                 
-                # Gunakan gambar yang sudah di-preprocessing untuk ditampilkan
-                preprocessed_img = preprocess_image_like_dataset(image, target_size=100)
+                # Untuk tampilan, resize ke 100x100
+                preprocessed_img = cv2.resize(np.array(image.convert('RGB')), (100, 100), interpolation=cv2.INTER_AREA)
                 preprocessed_pil = Image.fromarray(preprocessed_img)
                 
                 img_io = io.BytesIO()

requirements.txt

@@ -5,6 +5,4 @@ joblib
 scikit-image
 scikit-learn
 pillow
-opencv-python-headless
-rembg
-onnxruntime
+opencv-python-headless