首先dlib库进行人脸识别与特征标定获得人脸对应的68个特征点，然后通过嘴巴张开距离占面部识别框宽度的比例、眉毛上扬程度、眯眼睛程度对用户情绪进行判定

【“魔镜”系列】基于dlib的情绪识别

原理

首先dlib库进行人脸识别与特征标定获得人脸对应的68个特征点，然后通过嘴巴张开距离占面部识别框宽度的比例、眉毛上扬程度、眯眼睛程度对用户情绪进行判定
参考：python3+dlib人脸识别及情绪分析

代码

class EmotionRecognition:
    def __init__(self):
        self.detector = dlib.get_frontal_face_detector()            # 使用特征提取器get_frontal_face_detector
        self.predictor = dlib.shape_predictor(DATA_PATH)            # dlib的68点模型，使用训练好的特征预测器

    def learning_face(self, path):
        im_read = cv2.imread(path, 1)
        face_list = self.detector(cv2.cvtColor(im_read, cv2.COLOR_RGB2GRAY), 0)   # 对灰度图像使用检测器检测
        if len(face_list) != 0:                                     # 检测器检测到人脸
            for i in range(len(face_list)):                         # 获取68个特征点
                for k, d in enumerate(face_list):                   # 获取索引和数据，k为索引，d为对象
                    box_width = d.right() - d.left()                # 计算人脸框边长
                    appear = self.predictor(im_read, d)             # 获取68个特征点坐标
                    mouth_higth = (appear.part(66).y - appear.part(62).y) / box_width
                    # 眉毛直线拟合列表
                    brow_x = []
                    brow_y = []
                    for j in range(17, 21):
                        brow_x.append(appear.part(j).x)
                        brow_y.append(appear.part(j).y)
                    z1 = np.polyfit(np.array(brow_x), np.array(brow_y), 1)
                    brow_k = -round(z1[0], 3)                       # 拟合成一条直线
                    eye_hight = ((appear.part(41).y - appear.part(37).y + appear.part(40).y - appear.part(38).y +
                               appear.part(47).y - appear.part(43).y + appear.part(46).y - appear.part(44).y)
                                 / 4) / box_width                   # 获取眼睛睁开大小
                    upload_qiniu(path)                              # 上传至图床
                    if round(mouth_higth >= 0.03):
                        if eye_hight >= 0.056:
                            send_face_data_save(path, "3")          # 惊讶
                            return json.dumps({"state": "success", "result": "amazing"})
                        else:
                            send_face_data_save(path, "4")          # 开心
                            return json.dumps({"state": "success", "result": "happy"})
                    else:
                        if brow_k <= -0.3:
                            send_face_data_save(path, "1")          # 生气
                            return json.dumps({"state": "success", "result": "angry"})
                        else:
                            send_face_data_save(path, "2")          # 正常
                            return json.dumps({"state": "success", "result": "nature"})
                break                                               # 只识别一张人脸
        return json.dumps({"state": "error", "result": "no face"})  # 未识别到人脸

此为博主副博客，留言请去主博客，转载请注明出处：https://www.baby7blog.com/myBlog/75.html