Update app.py

app.py

@@ -1,62 +1,52 @@
 import gradio as gr
-from transformers import pipeline
-from PIL import Image
+from PIL import Image, ImageStat, ExifTags
 import numpy as np
+import cv2
 
 # ----------------------------
-# Inisialisasi model publik
+# Fungsi hitung sharpness / blur
 # ----------------------------
-model1 = pipeline("image-classification", model="microsoft/resnet-50")
-model2 = pipeline("image-classification", model="google/vit-base-patch16-224")
+def calculate_blur(image):
+    image_cv = np.array(image.convert("L"))  # konversi ke grayscale
+    laplacian_var = cv2.Laplacian(image_cv, cv2.CV_64F).var()
+    return laplacian_var
 
 # ----------------------------
-# Fungsi deteksi dengan ensemble tanpa "Tidak Pasti"
+# Fungsi deteksi metadata
 # ----------------------------
-def detect_image(image):
-    results = []
-
-    # Model 1
-    res1 = model1(image)[0]
-    label1, score1 = res1['label'].lower(), res1['score']
-    label1_final = 'human' if 'person' in label1 or 'human' in label1 else 'ai'
-    results.append((label1_final, score1))
-
-    # Model 2
-    res2 = model2(image)[0]
-    label2, score2 = res2['label'].lower(), res2['score']
-    label2_final = 'human' if 'person' in label2 or 'human' in label2 else 'ai'
-    results.append((label2_final, score2))
-
-    # ----------------------------
-    # Voting atau ambil confidence tertinggi
-    # ----------------------------
-    votes = [r[0] for r in results]
-    if votes[0] == votes[1]:
-        final_label = votes[0]  # mayoritas sama
-        final_conf = np.mean([r[1] for r in results if r[0]==final_label])*100
-    else:
-        # ambil model dengan confidence tertinggi
-        if results[0][1] > results[1][1]:
-            final_label = results[0][0]
-            final_conf = results[0][1]*100
-        else:
-            final_label = results[1][0]
-            final_conf = results[1][1]*100
+def has_camera_exif(image):
+    try:
+        exif = image._getexif()
+        if exif:
+            for tag, value in exif.items():
+                decoded = ExifTags.TAGS.get(tag, tag)
+                if decoded in ["Make", "Model"]:
+                    return True
+    except:
+        return False
+    return False
 
-    # ----------------------------
-    # Hasil akhir
-    # ----------------------------
-    if final_label == 'human':
-        final_result = f"✅ Foto Asli ({final_conf:.2f}%)"
+# ----------------------------
+# Fungsi prediksi AI vs Real
+# ----------------------------
+def detect_image(image):
+    blur_score = calculate_blur(image)
+    exif_present = has_camera_exif(image)
+    
+    # Threshold empiris (bisa disesuaikan)
+    # blur rendah + metadata → Foto Asli
+    # blur tinggi + tidak ada metadata → AI
+    if exif_present or blur_score > 100:
+        final_result = f"✅ Foto Asli"
     else:
-        final_result = f"🤖 AI Detected ({final_conf:.2f}%)"
-
+        final_result = f"🤖 AI Detected"
+    
     output = f"""
 ### Hasil Deteksi:
 {final_result}
 
-**Model 1 (ResNet-50):** {label1} ({score1*100:.2f}%)
-**Model 2 (ViT):** {label2} ({score2*100:.2f}%)
+**Blur Score:** {blur_score:.2f}
+**Metadata Kamera:** {'Ada' if exif_present else 'Tidak Ada'}
 """
     return output
 
@@ -67,8 +57,8 @@ iface = gr.Interface(
     fn=detect_image,
     inputs=gr.Image(type="pil"),
     outputs="markdown",
-    title="AI vs Foto Asli Detector",
-    description="Unggah gambar, sistem akan mendeteksi apakah gambar AI atau Foto Asli berdasarkan 2 model publik."
+    title="AI vs Foto Asli Detector (Gratis)",
+    description="Unggah gambar, sistem akan mendeteksi apakah gambar kemungkinan besar asli atau dihasilkan AI berdasarkan analisis blur dan metadata."
 )
 
 if __name__ == "__main__":