---

0 前言

人脸表情识别一直都是个热门的课题，今天学长向大家介绍如何使用深度学习技术进行人脸表情识别。

---

1 技术介绍

1.1 技术概括

面部表情识别技术源于1971年心理学家Ekman和Friesen的一项研究，他们提出人类主要有六种基本情感，每种情感以唯一的表情来反映当时的心理活动，这六种情感分别是愤怒（anger）、高兴(happiness)、悲伤 (sadness)、惊讶(surprise)、厌恶(disgust)和恐惧(fear)。

尽管人类的情感维度和表情复杂度远不是数字6可以量化的，但总体而言，这6种也差不多够描述了。

1.2 目前表情识别实现技术

2 实现效果

废话不多说，先上实现效果

3 深度学习表情识别实现过程

3.1 网络架构

面部表情识别CNN架构（改编自 埃因霍芬理工大学PARsE结构图）

其中，通过卷积操作来创建特征映射，将卷积核挨个与图像进行卷积，从而创建一组要素图，并在其后通过池化（pooling）操作来降维。

3.2 数据

主要来源于kaggle比赛，下载地址。
有七种表情类别： (0=Angry, 1=Disgust, 2=Fear, 3=Happy, 4=Sad, 5=Surprise, 6=Neutral).
数据是48x48 灰度图，格式比较奇葩。
第一列是情绪分类，第二列是图像的numpy，第三列是train or test。

3.3 实现流程

3.4 部分实现代码

import cv2
import sys
import json
import numpy as np
from keras.models import model_from_json


emotions = ['angry', 'fear', 'happy', 'sad', 'surprise', 'neutral']
cascPath = sys.argv[1]

faceCascade = cv2.CascadeClassifier(cascPath)
noseCascade = cv2.CascadeClassifier(cascPath)


# load json and create model arch
json_file = open('model.json','r')
loaded_model_json = json_file.read()
json_file.close()
model = model_from_json(loaded_model_json)

# load weights into new model
model.load_weights('model.h5')

# overlay meme face
def overlay_memeface(probs):
    if max(probs) > 0.8:
        emotion = emotions[np.argmax(probs)]
        return 'meme_faces/{}-{}.png'.format(emotion, emotion)
    else:
        index1, index2 = np.argsort(probs)[::-1][:2]
        emotion1 = emotions[index1]
        emotion2 = emotions[index2]
        return 'meme_faces/{}-{}.png'.format(emotion1, emotion2)

def predict_emotion(face_image_gray): # a single cropped face
    resized_img = cv2.resize(face_image_gray, (48,48), interpolation = cv2.INTER_AREA)
    # cv2.imwrite(str(index)+'.png', resized_img)
    image = resized_img.reshape(1, 1, 48, 48)
    list_of_list = model.predict(image, batch_size=1, verbose=1)
    angry, fear, happy, sad, surprise, neutral = [prob for lst in list_of_list for prob in lst]
    return [angry, fear, happy, sad, surprise, neutral]

video_capture = cv2.VideoCapture(0)
while True:
    # Capture frame-by-frame
    ret, frame = video_capture.read()

    img_gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY,1)


    faces = faceCascade.detectMultiScale(
        img_gray,
        scaleFactor=1.1,
        minNeighbors=5,
        minSize=(30, 30),
        flags=cv2.cv.CV_HAAR_SCALE_IMAGE
    )

    # Draw a rectangle around the faces
    for (x, y, w, h) in faces:

        face_image_gray = img_gray[y:y+h, x:x+w]
        filename = overlay_memeface(predict_emotion(face_image_gray))

        print filename
        meme = cv2.imread(filename,-1)
        # meme = (meme/256).astype('uint8')
        try:
            meme.shape[2]
        except:
            meme = meme.reshape(meme.shape[0], meme.shape[1], 1)
        # print meme.dtype
        # print meme.shape
        orig_mask = meme[:,:,3]
        # print orig_mask.shape
        # memegray = cv2.cvtColor(orig_mask, cv2.COLOR_BGR2GRAY)
        ret1, orig_mask = cv2.threshold(orig_mask, 10, 255, cv2.THRESH_BINARY)
        orig_mask_inv = cv2.bitwise_not(orig_mask)
        meme = meme[:,:,0:3]
        origMustacheHeight, origMustacheWidth = meme.shape[:2]

        roi_gray = img_gray[y:y+h, x:x+w]
        roi_color = frame[y:y+h, x:x+w]

        # Detect a nose within the region bounded by each face (the ROI)
        nose = noseCascade.detectMultiScale(roi_gray)

        for (nx,ny,nw,nh) in nose:
            # Un-comment the next line for debug (draw box around the nose)
            #cv2.rectangle(roi_color,(nx,ny),(nx+nw,ny+nh),(255,0,0),2)

            # The mustache should be three times the width of the nose
            mustacheWidth =  20 * nw
            mustacheHeight = mustacheWidth * origMustacheHeight / origMustacheWidth

            # Center the mustache on the bottom of the nose
            x1 = nx - (mustacheWidth/4)
            x2 = nx + nw + (mustacheWidth/4)
            y1 = ny + nh - (mustacheHeight/2)
            y2 = ny + nh + (mustacheHeight/2)

            # Check for clipping
            if x1 < 0:
                x1 = 0
            if y1 < 0:
                y1 = 0
            if x2 > w:
                x2 = w
            if y2 > h:
                y2 = h


            # Re-calculate the width and height of the mustache image
            mustacheWidth = (x2 - x1)
            mustacheHeight = (y2 - y1)

            # Re-size the original image and the masks to the mustache sizes
            # calcualted above
            mustache = cv2.resize(meme, (mustacheWidth,mustacheHeight), interpolation = cv2.INTER_AREA)
            mask = cv2.resize(orig_mask, (mustacheWidth,mustacheHeight), interpolation = cv2.INTER_AREA)
            mask_inv = cv2.resize(orig_mask_inv, (mustacheWidth,mustacheHeight), interpolation = cv2.INTER_AREA)

            # take ROI for mustache from background equal to size of mustache image
            roi = roi_color[y1:y2, x1:x2]

            # roi_bg contains the original image only where the mustache is not
            # in the region that is the size of the mustache.
            roi_bg = cv2.bitwise_and(roi,roi,mask = mask_inv)

            # roi_fg contains the image of the mustache only where the mustache is
            roi_fg = cv2.bitwise_and(mustache,mustache,mask = mask)

            # join the roi_bg and roi_fg
            dst = cv2.add(roi_bg,roi_fg)

            # place the joined image, saved to dst back over the original image
            roi_color[y1:y2, x1:x2] = dst

            break

    #     cv2.rectangle(frame, (x, y), (x+w, y+h), (0, 255, 0), 2)
    #     angry, fear, happy, sad, surprise, neutral = predict_emotion(face_image_gray)
    #     text1 = 'Angry: {}     Fear: {}   Happy: {}'.format(angry, fear, happy)
    #     text2 = '  Sad: {} Surprise: {} Neutral: {}'.format(sad, surprise, neutral)
    #
    # cv2.putText(frame, text1, (50, 50), cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 0, 0), 3)
    # cv2.putText(frame, text2, (50, 150), cv2.FONT_HERSHEY_SIMPLEX, 2, (255, 0, 0), 3)

    # Display the resulting frame
    cv2.imshow('Video', frame)

    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

# When everything is done, release the capture
video_capture.release()
cv2.destroyAllWindows()

4 最后