"""

Evaluate the performance of face sketch synthesis.

Method: PCA and SVM face recognition.

Modified from face recognition examples from sklearn:

http://scikit-learn.org/stable/tutorial/statistical_inference/putting_together.html

"""

from __future__ import print_function

from time import time

from sklearn.decomposition import PCA

import numpy as np

import cv2 as cv

import os

def load_dataset(dir_path, img_list=None, size=(200, 250)):

data = []

label = []

for root, dirs, files in os.walk(dir_path):

if img_list:

files = img_list

for f in sorted(files):

img = cv.imread(os.path.join(root, f), 0)

if img.shape != size[::-1]:

out_size = np.array(size[::-1])

border = out_size - img.shape

border1 = np.floor(border / 2.).astype('int')

border2 = np.ceil(border / 2.).astype('int')

img = cv.copyMakeBorder(img, border1[0], border2[0], border1[1], border2[1], cv.BORDER_CONSTANT, value=255)

data.append(img)

if f.startswith('AR'):

name = f[3:-4]

else:

name = f[:-4]

label.append(int(name))

return np.array(data), np.array(label)

def pred_acc(train_pca, y_train, test_pca, y_test, topk=1):

pred = []

# train_pca = train_pca / np.linalg.norm(train_pca)

# test_pca = test_pca / np.linalg.norm(test_pca)

for i in test_pca:

dist = np.sum((i - train_pca)**2, 1)

# dist = 1 - np.dot(train_pca, i)

pred.append(np.argmin(dist))

pred = np.array(pred)

pred_label = y_train[pred]

return (np.sum(pred_label==y_test)*1. / y_test.shape[0])

def PCA_recognition(train_dir, test_dirs, n_components=30, test_file_lists=None):

X_train, y_train = load_dataset(train_dir)

n_samples, h, w = X_train.shape

X_train = X_train.reshape(-1, h*w)

mean_face = np.mean(X_train, 0)

# X_train = X_train - mean_face

pca = PCA(n_components=n_components, svd_solver='auto', whiten=True).fit(X_train)

for idx, test_dir in enumerate(test_dirs):

if test_file_lists:

X_test, y_test = load_dataset(test_dir, test_file_lists[idx])

else:

X_test, y_test = load_dataset(test_dir)

X_test = X_test.reshape(-1, h*w)

# X_test = X_test - mean_face

X_train_pca = pca.transform(X_train)

X_test_pca = pca.transform(X_test)

print(test_dir.split('/')[-1], '\t', pred_acc(X_train_pca, y_train, X_test_pca, y_test))

if __name__ == '__main__':

print('Evaluate CUHK dataset')

train_dir = './Data/CUHK_student_test/sketches'

test_dirs = ['./other_results/CUHK/MRF',

'./other_results/CUHK/MWF',

'./other_results/CUHK/SSD',

'./other_results/CUHK/FCNN',

'./other_results/CUHK/BFCN',

'./result_CUHK/sketch']

PCA_recognition(train_dir, test_dirs, 98)

print('Evaluate AR dataset')

train_dir = './Data/AR_test/sketches'

test_dirs = ['./other_results/AR/MRF',

'./other_results/AR/MWF',

'./other_results/AR/SSD',

'./other_results/AR/FCNN',

'./other_results/AR/BFCN',

'./result_AR/sketch']

img_list =[x.strip() for x in open('./Data/AR/test.txt').readlines()]

test_file_lists = [img_list, img_list, img_list, img_list, img_list, None]

PCA_recognition(train_dir, test_dirs, 10, test_file_lists)