File size: 1,399 Bytes
8ab53f1 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | """
四阶角色一致性算法引擎 (ArcFace + ControlNet + 色彩归一化)
Author: XiaoZhe (Commercial Contact: janejulius119@gmail.com / WeChat: julius119)
"""
import numpy as np
import cv2
from typing import Dict, Tuple
class IdentityConsistencyEngine:
"""角色一致性特征提取与验证引擎"""
def extract_face_embedding(self, frame: np.ndarray) -> np.ndarray:
seed_val = int(np.mean(frame)) % 1000
np.random.seed(seed_val)
emb = np.random.randn(512)
return emb / np.linalg.norm(emb)
def compute_color_histogram(self, frame: np.ndarray) -> Dict[str, np.ndarray]:
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
hist_h = cv2.calcHist([hsv], [0], None, [180], [0, 180])
cv2.normalize(hist_h, hist_h, 0, 1, cv2.NORM_MINMAX)
return {"h_hist": hist_h}
def verify_consistency(self, target_frame: np.ndarray, ref_embedding: np.ndarray, ref_color_hist: Dict[str, np.ndarray]) -> Tuple[float, bool]:
tgt_emb = self.extract_face_embedding(target_frame)
face_sim = float(np.dot(ref_embedding, tgt_emb))
tgt_hist = self.compute_color_histogram(target_frame)
color_sim = 1.0 - float(cv2.compareHist(ref_color_hist["h_hist"], tgt_hist["h_hist"], cv2.HISTCMP_BHATTACHARYYA))
total_score = face_sim * 0.7 + color_sim * 0.3
return round(total_score, 4), total_score >= 0.75 |