采用 OpenCV DNN 模块实现的人脸性别及年龄检测，整个项目比较简单、清晰明了，过程主要包括：

[1] - 检测图片中的人脸框(如，采用 dlib 库).

[2] - 检测人脸的性别和年龄.

[3] - 检测结果可视化.

1. 相关模型

所采用的模型分别如下：

[1] - 人脸框检测模型

这里采用了 dlib 的 CNN 人脸框检测模型：

或者，

OpenCV DNN 模块自带的残差网络的人脸检测模型：

[2] - 性别检测模型(Caffe)：

性别共两类：Male，Female

[3] - 年龄预测模型(Caffe)：

年龄共八类：(0-2), (4-6), (8-12), (15-20), (25-32), (38-43), (48-53), (60-100)

2. 具体实现

#!--*-- coding:utf-8 --*--
import os
import cv2
import time
import matplotlib.pyplot as plt
import dlib

class AgeGenderDetector():

    def __init__(self):
        # Load model and networks
        # Age
        self.ageList = ['(0-2)', 
                        '(4-6)', 
                        '(8-12)', 
                        '(15-20)', 
                        '(25-32)', 
                        '(38-43)', 
                        '(48-53)', 
                        '(60-100)']
        ageProto = "age_deploy.prototxt"
        ageModel = "age_net.caffemodel"
        self.ageNet = cv2.dnn.readNet(ageModel, ageProto)

        # Gender
        self.genderList = ['Male', 'Female']
        genderProto = "gender_deploy.prototxt"
        genderModel = "gender_net.caffemodel"
        self.genderNet = cv2.dnn.readNet(genderModel, genderProto)

        # Face detection - OpenCV DNN 
        faceProto = "opencv_face_detector.pbtxt"
        faceModel = "opencv_face_detector_uint8.pb"
        self.faceNet = cv2.dnn.readNet(faceModel, faceProto)

        # Face detection - dlib
        dlib_cnn_face_model = "mmod_human_face_detector.dat"
        self.dlib_cnn_face_detector = 
            dlib.cnn_face_detection_model_v1(dlib_cnn_face_model)

        self.mean_values = (78.4263377603, 87.7689143744, 114.895847746)
        self.padding = 20


    def dlib_cnn_face_detect(self, img):
        detections = self.dlib_cnn_face_detector(img, 1)
        bboxes = []
        if len(detections) > 0:
            for i, d in enumerate(detections):
                x1 = d.rect.left()
                y1 = d.rect.top()
                x2 = d.rect.right()
                y2 = d.rect.bottom()
                bboxes.append([x1, y1, x2, y2]) #

        return bboxes


    def opencv_dnn_face_detect(self, img_cv2, conf_threshold=0.7):
        height, width, _ = img_cv2.shape
        blob = cv2.dnn.blobFromImage(img_cv2, 
                                     1.0, 
                                     (300, 300), 
                                     [104, 117, 123], 
                                     True, 
                                     False)

        self.faceNet.setInput(blob)
        detections = self.faceNet.forward()
        bboxes = []
        for i in range(detections.shape[2]):
            confidence = detections[0, 0, i, 2]
            if confidence > conf_threshold:
                x1 = int(detections[0, 0, i, 3] * width)
                y1 = int(detections[0, 0, i, 4] * height)
                x2 = int(detections[0, 0, i, 5] * width)
                y2 = int(detections[0, 0, i, 6] * height)
                bboxes.append([x1, y1, x2, y2]) #
        return bboxes


    def detect(self, img_file):
        start = time.time()
        img_cv2 = cv2.imread(img_file)
        img_rgb = cv2.cvtColor(img_cv2, cv2.COLOR_BGR2RGB)

        # Face box detection
        # face_bboxes = self.opencv_dnn_face_detect(img_cv2)
        face_bboxes = self.dlib_cnn_face_detect(img_rgb) # RGB order
        if not face_bboxes:
            print("No face Detected, Checking next image")

        for bbox in face_bboxes:
            # print(bbox)
            # single face 
            face = img_cv2[
                max(0, bbox[1]-self.padding):
                min(bbox[3] + self.padding, img_cv2.shape[0]-1),
                max(0, bbox[0]-self.padding):
                min(bbox[2] + self.padding, img_cv2.shape[1]-1)] 

            # Gender
            blob = cv2.dnn.blobFromImage(face, 1.0, (227, 227), 
                                         self.mean_values, swapRB=False)
            self.genderNet.setInput(blob)
            genderPreds = self.genderNet.forward()
            gender = self.genderList[genderPreds[0].argmax()]
            # print("Gender Output : {}".format(genderPreds))
            print("Gender : {}, conf = {:.3f}".format(
                gender, genderPreds[0].max()))

            # Age
            self.ageNet.setInput(blob)
            agePreds = self.ageNet.forward()
            age = self.ageList[agePreds[0].argmax()]
            print("Age Output : {}".format(agePreds))
            print("Age : {}, conf = {:.3f}".format(age, agePreds[0].max()))

            # vis
            label = "{}, {}".format(gender, age)
            cv2.rectangle(img_cv2, 
                          (bbox[0], bbox[1]), (bbox[2], bbox[3]), 
                          (0, 255, 0), 
                          int(round(img_cv2.shape[0]/150)), 
                          8)
            cv2.putText(img_cv2, 
                        label, 
                        (bbox[0], bbox[1]-10), 
                        cv2.FONT_HERSHEY_SIMPLEX, 
                        0.6, 
                        (0, 255, 255), 
                        2, 
                        cv2.LINE_AA)

        # cv2.imwrite("age-gender-out-{}".format(args.input),frameFace)
        print("time : {:.3f}".format(time.time() - start))
        plt.figure(figsize=(10, 8))
        plt.axis("off")
        plt.imshow(img_cv2[:,:,::-1])
        plt.show()

        return True


if __name__ == '__main__':
    print("[INFO] Starting AgeGender Demo...")

    img_path = "/path/to/images/"
    img_files = [os.path.join(img_path, img_name) 
                 for img_name in os.listdir(img_path)]
    age_gender_detector = AgeGenderDetector()

    for img_file in img_files:
        print("Processing file: {}".format(img_file))
        age_gender_detector.detect(img_file)

    print("[INFO]Done.")

输入如：

Gender : Female, conf = 0.984
Age : (25-32), conf = 0.500
time : 0.474

image

Gender : Female, conf = 0.999
Age : (25-32), conf = 0.408
time : 0.211

image

3. 参考材料

主要参考：

[1] - Age and Gender Classification Using Convolutional Neural Network

[2] - Github - spmallick/learnopencv/AgeGender

[3] - 论文阅读 - 基于CNN的年龄和性别分类 - AIUAI

Last modification：April 23rd, 2019 at 09:02 pm
© 允许规范转载