Implement image processing function to remove background and resize images in preprocessing pipeline

app.py

@@ -6,6 +6,7 @@ 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__)
 
@@ -41,6 +42,48 @@ CLASS_NAMES = [
     "Salak",
 ]
 
+def process_image(pil_img, target_size=(100, 100)):
+    img_io = io.BytesIO()
+    pil_img.save(img_io, format='PNG')
+    img_bytes = img_io.getvalue()
+    
+    output_array = remove(img_bytes)
+    nparr = np.frombuffer(output_array, np.uint8)
+    img_rgba = cv2.imdecode(nparr, cv2.IMREAD_UNCHANGED)
+    
+    alpha = img_rgba[:, :, 3]
+    coords = cv2.findNonZero(alpha)
+    
+    if coords is None:
+        return np.array(pil_img.resize(target_size).convert("RGB"))
+    
+    x, y, w, h = cv2.boundingRect(coords)
+    cropped_img = img_rgba[y:y+h, x:x+w]
+    
+    old_h, old_w = cropped_img.shape[:2]
+    desired_w, desired_h = target_size
+    
+    ratio = min(desired_w / old_w, desired_h / old_h)
+    new_w = int(old_w  * ratio)
+    new_h = int(old_h * ratio)
+    
+    resized_img = cv2.resize(cropped_img, (new_w, new_h), interpolation=cv2.INTER_AREA)
+    
+    final_bg = np.zeros((desired_h, desired_w, 4), dtype=np.uint8)
+    
+    x_offset = (desired_w - new_w) // 2
+    y_offset = (desired_h - new_h) // 2
+    final_bg[y_offset:y_offset+new_h, x_offset:x_offset+new_w] = resized_img
+    
+    white_bg = np.ones((desired_h, desired_w, 3), dtype=np.uint8) * 255
+    
+    alpha_channel = final_bg[:, :, 3] / 255.0
+    
+    for c in range(3):
+        white_bg[:, :, c] = (final_bg[:, :, c] * alpha_channel + white_bg[:, :, c] * (1.0 - alpha_channel))
+        
+    return white_bg.astype(np.uint8)
+
 
 def extract_color_histogram(img, bins=(8, 8, 8)):
     hist = cv2.calcHist([img], [0, 1, 2], None, bins, [0, 256, 0, 256, 0, 256])
@@ -58,13 +101,9 @@ def extract_lbp_features(gray_img, P=8, R=1, method="uniform"):
 
 
 def preprocessing_pipeline(pil_img):
-    img = np.array(pil_img)
-
-    # ubah format dari RGBA menjadi RGB
-    if img.shape[-1] == 4:
-        img = img[:, :, :3]
+    clean_img_rgb = process_image(pil_img)
 
-    img_float = img.astype(np.float32) / 255.0
+    img_float = clean_img_rgb.astype(np.float32) / 255.0
     img_uint8 = (img_float * 255).astype(np.uint8)
 
     feat_color = extract_color_histogram(img_uint8)