Update emotion.py

emotion.py

@@ -23,6 +23,12 @@ ZOOM_FACTOR = 1.5
 DETECTOR_BACKEND = "mtcnn"     # 想快一點可改 "opencv"
 ALIGN_FACE = True
 
+# ✅ 新增：防止背景誤判為臉（你要的核心）
+# 臉框面積佔整張畫面比例下限（太小通常是誤判）
+MIN_FACE_AREA_RATIO = 0.05   # 5%（可調：0.03~0.08）
+# 偵測器信心值下限（MTCNN/RetinaFace 通常有 confidence）
+MIN_DET_CONF = 0.90          # 可調：0.80~0.95
+
 
 # =========================
 # 情緒中文 + 分數映射（7 -> 5）
@@ -38,7 +44,6 @@ EMO_ZH = {
     "unknown": "未知",
 }
 
-# 你要的：把 7 種情緒塞進 5 分心情分數
 EMO_TO_SCORE = {
     "angry": 1,
     "disgust": 1,
@@ -92,13 +97,13 @@ class AppState:
 
     final_score: int = 3
 
-    # ✅ 新增：最近有沒有臉（避免節流期間一直紅框）
+    # 最近有沒有臉（避免節流期間一直紅框）
     last_face_t: float = 0.0
     face_in_box: bool = False
 
 
 # =========================
-# UI HTML (強制黑色文字)
+# UI HTML
 # =========================
 def _hint_html(msg: str) -> str:
     return f"""
@@ -107,22 +112,18 @@ def _hint_html(msg: str) -> str:
     </div>
     """
 
-
 def _camera_hint_before_start() -> str:
     return _hint_html(
         "請先按「Click to access webcam / 允許」開啟攝影機；"
         "接著，請按「錄製（Record）」開始。"
     )
 
-
 def _camera_hint_done() -> str:
     return _hint_html("已完成辨識，可按「重新攝影機辨識」再次進行辨識。")
 
-
 def _camera_hint_stopped() -> str:
     return _hint_html("已停止辨識。可按「重新攝影機辨識」再試一次。")
 
-
 def _result_card_html(emo_key: str, conf: Optional[float]) -> str:
     zh = EMO_ZH.get(emo_key, emo_key)
     score = EMO_TO_SCORE.get(emo_key, 3)
@@ -135,7 +136,7 @@ def _result_card_html(emo_key: str, conf: Optional[float]) -> str:
         padding:20px;
         border:1px solid #E6E6E6;
         background:#FFFFFF;
-        color:#000000; 
+        color:#000000;
     ">
       <div style="font-size:14px;color:#000000;margin-bottom:8px;">
         辨識結果
@@ -158,13 +159,15 @@ def _result_card_html(emo_key: str, conf: Optional[float]) -> str:
     </div>
     """
 
-
 def _score_radio_value(score: int) -> str:
     score = int(score)
     score = min(5, max(1, score))
     return SCORE_LABELS[score]
 
 
+# =========================
+# 影像工具
+# =========================
 def _downsample_rgb(frame: np.ndarray, target_w: int) -> np.ndarray:
     if frame is None:
         return None
@@ -177,11 +180,9 @@ def _downsample_rgb(frame: np.ndarray, target_w: int) -> np.ndarray:
     img = img.resize((target_w, new_h))
     return np.array(img)
 
-
 def _zoom_center(frame: np.ndarray, zoom: float) -> np.ndarray:
     if frame is None or zoom <= 1.0:
         return frame
-
     h, w = frame.shape[:2]
     new_w = int(w / zoom)
     new_h = int(h / zoom)
@@ -192,7 +193,7 @@ def _zoom_center(frame: np.ndarray, zoom: float) -> np.ndarray:
 
 
 # =========================
-# 引導框（臉要放哪裡）
+# 引導框
 # =========================
 GUIDE_W_RATIO = 0.55
 GUIDE_H_RATIO = 0.70
@@ -202,10 +203,8 @@ def _guide_box(frame: np.ndarray) -> Tuple[int, int, int, int]:
     h, w = frame.shape[:2]
     gw = int(w * GUIDE_W_RATIO)
     gh = int(h * GUIDE_H_RATIO)
-
     cx = w // 2
     cy = int(h * (0.5 + GUIDE_Y_OFFSET))
-
     x1 = max(0, cx - gw // 2)
     y1 = max(0, cy - gh // 2)
     x2 = min(w - 1, cx + gw // 2)
@@ -227,8 +226,6 @@ def _draw_guide_overlay(frame_rgb: np.ndarray, detected_in_box: bool) -> np.ndar
     draw = ImageDraw.Draw(img)
 
     x1, y1, x2, y2 = _guide_box(frame_rgb)
-
-    # 紅/綠框
     color = (0, 200, 0) if detected_in_box else (220, 0, 0)
 
     h, w = frame_rgb.shape[:2]
@@ -264,7 +261,6 @@ def _iou(a: Tuple[int, int, int, int], b: Tuple[int, int, int, int]) -> float:
     union = aw * ah + bw * bh - inter
     return inter / max(1e-6, union)
 
-
 def _pick_face_by_tracking(
     faces: List[Dict],
     prev_bbox: Optional[Tuple[int, int, int, int]]
@@ -314,12 +310,40 @@ def _reset_state_for_camera(st: AppState) -> AppState:
     st.final_conf = None
     st.final_score = 3
 
-    # ✅ 新增欄位也要 reset
     st.last_face_t = 0.0
     st.face_in_box = False
     return st
 
 
+# =========================
+# ✅ 判斷「這個 face box 算不算真的臉」
+# =========================
+def _is_real_face(
+    pick: Dict,
+    bbox_full_xywh: Tuple[float, float, float, float],
+    frame_hw: Tuple[int, int],
+) -> bool:
+    fh, fw = frame_hw
+    x, y, w, h = bbox_full_xywh
+
+    # 1) 面積門檻（避免小塊背景誤判）
+    area_ratio = (w * h) / max(1.0, float(fw * fh))
+    if area_ratio < MIN_FACE_AREA_RATIO:
+        return False
+
+    # 2) 偵測 confidence（如果偵測器有提供）
+    # DeepFace.extract_faces 的 dict 常見會有 "confidence"（依 backend 而異）
+    det_conf = pick.get("confidence", None)
+    if det_conf is not None:
+        try:
+            if float(det_conf) < MIN_DET_CONF:
+                return False
+        except Exception:
+            pass
+
+    return True
+
+
 # =========================
 # Camera streaming
 # =========================
@@ -359,16 +383,12 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
 
     now = time.time()
 
-    # ✅ 節流時沿用最近一次「臉是否在框內」狀態（避免一直紅框）
+    # 節流：沿用最近一次「真的有人臉在框內」狀態
     detected_in_box_recent = st.face_in_box and (now - st.last_face_t) < 1.0
 
     if now - st.last_analyze_t < ANALYZE_EVERY_SEC:
         annotated = _draw_guide_overlay(frame_rgb, detected_in_box_recent)
-        if detected_in_box_recent:
-            result_html = _hint_html("已偵測到人臉，開始辨識情緒中…")
-        else:
-            result_html = _hint_html("未偵測到人臉，請把臉放進框內")
-
+        result_html = _hint_html("已偵測到人臉，開始辨識情緒中…" if detected_in_box_recent else "未偵測到人臉，請把臉放進框內")
         return (
             gr.update(visible=True, value=annotated),
             gr.update(value=result_html),
@@ -382,16 +402,20 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
 
     small = _downsample_rgb(frame_rgb, DOWNSAMPLE_W)
 
-    faces = DeepFace.extract_faces(
-        img_path=small,
-        detector_backend=DETECTOR_BACKEND,
-        enforce_detection=False,
-        align=ALIGN_FACE,
-    )
+    # ✅ 把「沒臉」當作 faces=[]（避免 backend 誤回傳奇怪結果）
+    try:
+        faces = DeepFace.extract_faces(
+            img_path=small,
+            detector_backend=DETECTOR_BACKEND,
+            enforce_detection=True,   # ✅ 變嚴格：沒偵測到臉就丟例外
+            align=ALIGN_FACE,
+        )
+    except Exception:
+        faces = []
 
     pick, bbox = _pick_face_by_tracking(faces, st.track_bbox)
 
-    # ✅ (A) 完全沒臉：紅框 + 不顯示情緒（避免「沒臉也顯示害怕」）
+    # (A) 完全沒臉：紅框 + 清掉狀態
     if pick is None or bbox is None or pick.get("face") is None:
         st.track_bbox = None
         st.face_in_box = False
@@ -408,7 +432,7 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
             gr.update(value="重新攝影機辨識"),
         )
 
-    # bbox 是 small 座標，換算回 frame_rgb
+    # bbox small -> frame
     sh, sw = small.shape[:2]
     fh, fw = frame_rgb.shape[:2]
     sx = fw / max(1, sw)
@@ -417,15 +441,31 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
     x, y, w, h = bbox
     bbox_full = (x * sx, y * sy, w * sx, h * sy)
 
+    # ✅ 新增：先檢查是不是「真的臉」（面積+confidence）
+    if not _is_real_face(pick, bbox_full, (fh, fw)):
+        st.track_bbox = None
+        st.face_in_box = False
+        st.last_face_t = now
+        st.stable_emo = "unknown"
+
+        annotated = _draw_guide_overlay(frame_rgb, False)
+        return (
+            gr.update(visible=True, value=annotated),
+            gr.update(value=_hint_html("未偵測到人臉，請把臉放進框內")),
+            gr.update(value=_score_radio_value(3)),
+            gr.update(value=_camera_hint_before_start()),
+            st,
+            gr.update(value="重新攝影機辨識"),
+        )
+
     guide = _guide_box(frame_rgb)
     detected_in_box = _bbox_center_in_box(bbox_full, guide)
 
-    # ✅ 更新「臉狀態」
     st.last_face_t = now
     st.face_in_box = detected_in_box
     st.track_bbox = bbox
 
-    # ✅ (B) 有臉但不在框內：紅框，不做情緒辨識
+    # (B) 有臉但不在框內：紅框，不做情緒辨識
     if not detected_in_box:
         annotated = _draw_guide_overlay(frame_rgb, False)
         return (
@@ -437,21 +477,38 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
             gr.update(value="重新攝影機辨識"),
         )
 
-    # ✅ (C) 有臉且在框內：綠框，才開始 analyze
+    # (C) 有臉且在框內：綠框，才開始 analyze
     face = pick["face"]
     if face.dtype != np.uint8:
         face = np.clip(face * 255.0, 0, 255).astype(np.uint8)
 
-    result = DeepFace.analyze(
-        img_path=face,
-        actions=["emotion"],
-        enforce_detection=False,
-    )
+    # ✅ analyze 也變嚴格：如果這塊其實不是臉，就當沒臉（避免誤判一路走到底）
+    try:
+        result = DeepFace.analyze(
+            img_path=face,
+            actions=["emotion"],
+            enforce_detection=True,
+        )
+    except Exception:
+        st.track_bbox = None
+        st.face_in_box = False
+        st.last_face_t = now
+        st.stable_emo = "unknown"
+
+        annotated = _draw_guide_overlay(frame_rgb, False)
+        return (
+            gr.update(visible=True, value=annotated),
+            gr.update(value=_hint_html("未偵測到人臉，請把臉放進框內")),
+            gr.update(value=_score_radio_value(3)),
+            gr.update(value=_camera_hint_before_start()),
+            st,
+            gr.update(value="重新攝影機辨識"),
+        )
+
     if isinstance(result, list):
         result = result[0]
     emo_dict = result.get("emotion", None)
 
-    # 若分析結果不完整：仍顯示綠框 + 辨識中提示
     if not isinstance(emo_dict, dict) or len(emo_dict) == 0:
         annotated = _draw_guide_overlay(frame_rgb, True)
         return (
@@ -466,7 +523,6 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
     top_key = max(emo_dict, key=lambda k: float(emo_dict[k]))
     top_conf = float(emo_dict[top_key])
 
-    # 信心不足：仍綠框 + 辨識中
     if top_conf < CONF_TH:
         annotated = _draw_guide_overlay(frame_rgb, True)
         return (
@@ -491,7 +547,6 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
     stable_key = st.emo_keys[stable_idx]
     st.stable_emo = stable_key
 
-    # 延遲確認
     if st.candidate_emo != stable_key:
         st.candidate_emo = stable_key
         st.candidate_since = now
@@ -533,7 +588,7 @@ def on_stream(frame_rgb: np.ndarray, st: AppState):
 
 
 # =========================
-# Mode / Buttons
+# Buttons
 # =========================
 def on_restart(st: AppState):
     st = _reset_state_for_camera(st)
@@ -546,7 +601,6 @@ def on_restart(st: AppState):
         gr.update(value="重新攝影機辨識"),
     )
 
-
 def on_stop(st: AppState):
     st.running = False
     st.finished = True