[PYTHON] [Failure] Find Maki Horikita by machine learning

【environment】

windows8.1 　Anaconda(python2.7)

【Overview】

I wrote the code to decide from the image whether it is Maki Horikita. Collect face image folders, digitize them and train them with scikit-learn.

Folder structure

|---face_category
　　　　　|---horikita_face(200 face image folders of Maki Horikita)
　　　　　|---joyu_face(200 face image folders other than Maki Horikita)
　　　　　|---horikita_siken(20 image folders of Maki Horikita for testing)
　　　　　|---face_rec2.py(horikita_Extract face from siken image)
　　　　　|---LinearSVC.py(Main processing)
　　　　　|---haarcascade_frontalface_alt.xml(Cascade file)

face_rec2.py

#-*- coding:utf-8 -*-
def picture_face(before_image, num):

    import cv2
    import os
    import sys
    
    cascade_path = "haarcascade_frontalface_alt.xml"
    
    #File reading
    image_file = cv2.imread('C:/Users/nobu/Desktop/my_programs/face_category/horikita_siken/%s' % before_image)
    
    #Grayscale conversion
    image_gray = cv2.cvtColor(image_file, cv2.COLOR_BGR2GRAY)
    
    #Acquire the features of the cascade classifier
    cascade = cv2.CascadeClassifier(cascade_path)
    
    #Execution of object recognition (face recognition)
    facerect = cascade.detectMultiScale(image_gray, scaleFactor=1.2, minNeighbors=2, minSize=(10, 10))
    
    if facerect == None:
        sys.exit()
    
    for rect in facerect:
        #Cut out only the face and save
        x = rect[0]
        y = rect[1]
        width = rect[2]
        height = rect[3]
        dst = image_file[y:y+height, x:x+width]
        resize_image = cv2.resize(dst,(256,256))
        new_image_path = 'C:/Users/nobu/Desktop/my_programs/face_category/' + str(num) + '_face.jpg'
        print new_image_path
        cv2.imwrite(new_image_path, resize_image)

LinearSVC.py

# -*- coding:utf-8 -*-

from sklearn.svm import LinearSVC
from PIL import Image
import os
import glob
import numpy as np
import face_rec2

current_dir = os.getcwd()

#Training data
#Convert Maki Horikita's image to numerical value
horikita_list = glob.glob(current_dir + "\\horikita_face\\*") 

horikita_dim2 = []

for image in horikita_list:
    #Exclude system files
    if image == current_dir + "\\horikita_face\\Thumbs.db":
        continue
    else:
        x = np.array([])
        horikita_dim3 = np.array(Image.open(image).convert('L'))
        #scikit-It seems that learn cannot handle more than 3 dimensions, so convert to 2 dimensions
        for i in xrange(256):
            x = np.r_[x,horikita_dim3[i]]
        horikita_dim2.append(x)

data_training1 = horikita_dim2

#Convert images other than Maki Horikita to numerical values
joyu_list = glob.glob(current_dir + "\\joyu_face\\*") 

joyu_dim2 = []

for image in joyu_list:
    #Exclude system files
    if image == current_dir + "\\joyu_face\\Thumbs.db":
        continue
    else:
        x = np.array([])
        joyu_dim3 = np.array(Image.open(image).convert('L'))
        #scikit-It seems that learn cannot handle more than 3 dimensions, so convert to 2 dimensions
        for i in xrange(256):
            x = np.r_[x,joyu_dim3[i]]
        joyu_dim2.append(x)
data_training2 = joyu_dim2

data_training = np.r_[data_training1, data_training2]

#Label setting
label_training = []
for i in xrange(400):
    if i < 200:
        label_training.append(1)
    else:
        label_training.append(0)

 #Learning
estimator = LinearSVC(C=0.5)
estimator.fit(data_training, label_training)

j = 0
#Test data
for i in xrange(1,21):
    face_rec2.picture_face(str(i) + '.jpg', i)
    horikita_siken = np.array([])
    try:
        horikita = np.array(Image.open('C:/Users/nobu/Desktop/my_programs/face_category/' + str(i) + '_face.jpg').convert('L'))
        for j in xrange(256):
                    horikita_siken = np.r_[horikita_siken,horikita[j]]
        data_test = horikita_siken
        
        #I'll predict
        label_prediction = estimator.predict(data_test)
        
        if label_prediction == 1:
            print "----------------------------------------------"
            print str(i) + ".jpg is Maki Horikita"
            print "----------------------------------------------"
        else:
            print "----------------------------------------------"
            print str(i) + ".jpg is not Maki Horikita"
            print "----------------------------------------------"
    except:
        print "----------------------------------------------"
        print str(i) + ".jpg could not be processed"
        print "----------------------------------------------"
        continue

【result】

Failure ...: weary: I tested with 20 face image folders of Maki Horikita for testing, but for some reason Some output says "It's not Maki Horikita". I would like to know if there are any solutions or improvements.

[Reference site]

SVM to try machine learning with scikit-learn Perform face recognition with OpenCV, trim and save only the face part [Python]