diff --git a/src/face_classifying.py b/src/face_classifying.py
new file mode 100644
index 0000000..cfb3545
--- /dev/null
+++ b/src/face_classifying.py
@@ -0,0 +1,28 @@
+from face_recognition import *
+
+def classify_faces(faces, threshold:float = 0.65):
+    
+    # classify each faces
+    res = {}
+    for face in faces:
+
+        score = 0        
+        max_score = 0
+        nearest_person = None
+        
+        for person, faces in res.items():
+            for face2 in faces:
+                score += is_same_face(face.img, face2.img)
+            
+            score /= len(faces)
+
+            if score > max_score:
+                nearest_person = person
+                max_score = score
+        
+        if max_score > threshold:
+            res[nearest_person].append(face)
+        else:
+            res[f"persone_{len(res)}"] = [face]
+
+    return res
\ No newline at end of file
diff --git a/src/face_recognition.py b/src/face_recognition.py
index e8e9df3..389502d 100644
--- a/src/face_recognition.py
+++ b/src/face_recognition.py
@@ -7,7 +7,7 @@ def extract_faces(faces_data:list, img:list, filename:str):
     
     recognizer = Recognizer.get()
     res = []
-    for i in range(1, len(faces_data[1])):
+    for i in range(len(faces_data[1])):
         face_data = faces_data[1][i]
 
         croped_img = recognizer.alignCrop(img, face_data)
@@ -33,4 +33,21 @@ def is_same_face(
     cosine_score = recognizer.match(faces1_features, faces2_features, cv2.FaceRecognizerSF_FR_COSINE)
     l2_score = recognizer.match(faces1_features, faces2_features, cv2.FaceRecognizerSF_FR_NORM_L2)
 
-    return cosine_score >= cosine_similarity_threshold and l2_score <= l2_similarity_threshold
\ No newline at end of file
+    return cosine_score >= cosine_similarity_threshold and l2_score <= l2_similarity_threshold
+
+def similarity_score(
+    face1_img:list, 
+    face2_img:list, 
+    cosine_similarity_threshold:float = 0.363,
+    l2_similarity_threshold:float = 1.128
+) -> bool:
+
+    recognizer = Recognizer.get()
+
+    faces1_features = recognizer.feature(face1_img)
+    faces2_features = recognizer.feature(face2_img)
+
+    cosine_score = recognizer.match(faces1_features, faces2_features, cv2.FaceRecognizerSF_FR_COSINE)
+    l2_score = recognizer.match(faces1_features, faces2_features, cv2.FaceRecognizerSF_FR_NORM_L2)
+
+    return (int(cosine_score >= cosine_similarity_threshold) + int(l2_score <= l2_similarity_threshold)) / 2
\ No newline at end of file
diff --git a/src/main.py b/src/main.py
index 8963a37..dd9c2ce 100644
--- a/src/main.py
+++ b/src/main.py
@@ -4,10 +4,13 @@ import os
 from utils import *
 from face_detection import *
 from face_recognition import *
+from face_classifying import *
 
 # list all files in data/raw
 files = list_all_files("./data/raw", ["jpg", "webp"])
 
+print(files)
+
 # process each files
 faces = []
 for file_path in files:
@@ -23,12 +26,23 @@ for file_path in files:
     # detect faces file
     faces_data = detect_face_in_image(img, scale=1)
 
-    print("faces_data: ", faces_data[1])
     # extract faces from file
-    faces += extract_faces(faces_data, img, filename)
+    extracted_faces = extract_faces(faces_data, img, filename)
 
-# classify each faces
-for face in faces:
-    face.save("./data/extracted")
+    faces += extracted_faces
+
+classified_faces = classify_faces(faces)
+
+# saving result by folders
+for person, faces in classified_faces.items():
+    dir_ = f"./data/results/{person}/"
+    if not os.path.exists(dir_):
+        os.makedirs(dir_)
+    
+    for face in faces:
+        face.save(dir_)
+
+
+print("classified_faces: ", classified_faces)
 
 cv2.destroyAllWindows()
\ No newline at end of file