utils.py

# from fileinput import filename
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import sklearn
import pickle
import cv2
# %matplotlib inline

#load all the prepared models
haar = cv2.CascadeClassifier('./model/haarcascade_frontalface_default.xml')
#pickle files
mean = pickle.load(open('./model/mean_preprocess.pickle','rb'))
model_svm = pickle.load(open('./model/model_svm.pickle','rb'))
model_pca = pickle.load(open('./model/pca_50.pickle','rb'))

print('Model loaded successfully')

#settings
gender_pre = ['Male','Female']
font = cv2.FONT_HERSHEY_COMPLEX

#model
def ml_pipeline_model(path,filename,color='bgr'):
    #stp1: read image in cv2
    img = cv2.imread(path)
    #stp2: convert into gray scale
    if color == 'bgr':
        gray = cv2.cvtColor(img,cv2.COLOR_BGR2GRAY)
    else:
        gray = cv2.cvtColor(img,cv2.COLOR_RGB2GRAY)
    #stp3: crop the face(using haar cascadeClassifier)
    faces = haar.detectMultiScale(gray,1.5,3)
    for x,y,w,h in faces:
        cv2.rectangle(img,(x,y),(x+w,y+h),(255,255,0),2) #drawing rectangle
        roi = gray[y:y+h,x:x+w] #crop image
        #stp4: normalization(0-1)
        roi = roi/255.0
        #stp5: resize images(100,100)
        if roi.shape[1]>100:
            roi_resize = cv2.resize(roi,(100,100),cv2.INTER_AREA)
        else:
            roi_resize = cv2.resize(roi,(100,100),cv2.INTER_CUBIC)
        #stp6: flattening image(1x10000)
        roi_reshape = roi_resize.reshape(1,10000) #1,-1
        #stp7: subtract from mean
        roi_mean = roi_reshape - mean
        #stp8: get eigen image
        eigen_image = model_pca.transform(roi_mean)
        #stp9: pass to ML model(svm)
        results = model_svm.predict_proba(eigen_image)[0]
        #stp10:
        predict = results.argmax() #0 or 1
        score = results[predict]
        #stp11:
        text = "%s : %0.2f"%(gender_pre[predict],score)
        cv2.putText(img,text,(x,y),font,1,(255,255,0),2)
    
    cv2.imwrite('./static/predict/{}'.format(filename),img)