Ror*_*ter 3 python opencv face-recognition computer-vision
我如何在OpenCV Python中进行性能测试来检查;
我在OpenCV中使用示例eigenface方法(来自Phillip - https://github.com/bytefish/facerecognition_guide),我只对结果感兴趣.如果有人能指出我正确的方向/展示例子,那将是很棒的.也许有一些我可以利用的功能？
byt*_*ish 22
首先很抱歉,回复花了这么长时间,但没有空余时间.实际上验证算法是一个非常有趣的话题,并且它真的不那么难.在这篇文章中,我将展示如何验证您的算法(我将采用FaceRecognizer,因为您已经要求它).和我的帖子一样,我将用一个完整的源代码示例来展示它,因为我认为用代码解释东西要容易得多.
所以每当人们告诉我"我的算法表现不好"时,我会问他们:
我的希望是,这篇文章将澄清一些混乱,并表明验证算法是多么容易.因为我从试验计算机视觉和机器学习算法中学到的是:
此帖子中的所有代码均归BSD许可,因此请随意将其用于您的项目.
任何计算机视觉项目最重要的任务之一是获取图像数据.您需要获得与生产中相同的图像数据,这样您在上线时就不会有任何不良体验.一个非常实用的示例:如果您想要识别野外的面部,那么在非常受控的场景中拍摄的图像上验证算法是没有用的.获取尽可能多的数据,因为数据是王道.那是为了数据.
一旦获得了一些数据并且编写了算法,就会对它进行评估.有几种验证策略,但我认为您应该从简单的交叉验证开始,然后从那里开始,有关交叉验证的信息,请参阅:
我们不会自己实现它,而是利用scikit - 学习一个伟大的开源项目:
它有一个非常好的文档和教程来验证算法:
所以计划如下:
cv2.FaceRecognizer成scikit-learn估算器.cv2.FaceRecognizer使用给定的验证和指标估算我们的绩效.首先,我想在要读取的图像数据上写一些单词,因为这个问题几乎总是会出现.为简单起见,我假设在示例中,图像(面部,您要识别的人)在文件夹中给出.每人一个文件夹.所以想象我有一个文件夹(数据集)调用images,子文件夹person1,person2等等:
philipp@mango:~/facerec/data/images$ tree -L 2 | head -n 20
.
|-- person1
| |-- 1.jpg
| |-- 2.jpg
| |-- 3.jpg
| |-- 4.jpg
|-- person2
| |-- 1.jpg
| |-- 2.jpg
| |-- 3.jpg
| |-- 4.jpg
[...]
其中一个公共可用数据集,即已经存在于这种文件夹结构中的是AT&T Facedatabase,可从以下位置获得:
一旦解压缩,它将看起来像这样(在我的文件系统上,它被解压缩到/home/philipp/facerec/data/at/,你的路径是不同的!):
philipp@mango:~/facerec/data/at$ tree .
.
|-- README
|-- s1
| |-- 1.pgm
| |-- 2.pgm
[...]
| `-- 10.pgm
|-- s2
| |-- 1.pgm
| |-- 2.pgm
[...]
| `-- 10.pgm
|-- s3
| |-- 1.pgm
| |-- 2.pgm
[...]
| `-- 10.pgm
...
40 directories, 401 files
首先,我们将定义一种read_images读取图像数据和标签的方法:
import os
import sys
import cv2
import numpy as np
def read_images(path, sz=None):
"""Reads the images in a given folder, resizes images on the fly if size is given.
Args:
path: Path to a folder with subfolders representing the subjects (persons).
sz: A tuple with the size Resizes
Returns:
A list [X,y]
X: The images, which is a Python list of numpy arrays.
y: The corresponding labels (the unique number of the subject, person) in a Python list.
"""
c = 0
X,y = [], []
for dirname, dirnames, filenames in os.walk(path):
for subdirname in dirnames:
subject_path = os.path.join(dirname, subdirname)
for filename in os.listdir(subject_path):
try:
im = cv2.imread(os.path.join(subject_path, filename), cv2.IMREAD_GRAYSCALE)
# resize to given size (if given)
if (sz is not None):
im = cv2.resize(im, sz)
X.append(np.asarray(im, dtype=np.uint8))
y.append(c)
except IOError, (errno, strerror):
print "I/O error({0}): {1}".format(errno, strerror)
except:
print "Unexpected error:", sys.exc_info()[0]
raise
c = c+1
return [X,y]
然后读取图像数据就像调用一样简单:
[X,y] = read_images("/path/to/some/folder")
因为某些算法(例如Eigenfaces,Fisherfaces)要求您的图像大小相等,所以我添加了第二个参数sz.通过传递元组sz,所有图像都会调整大小.因此,以下调用将调整所有图像/path/to/some/folder到100x100像素:
[X,y] = read_images("/path/to/some/folder", (100,100))
scikit-learn中的所有分类器都来自a BaseEstimator,它应该具有a fit和predict方法.该fit方法获取样本列表X和相应的标签y,因此映射到列车的方法是微不足道的cv2.FaceRecognizer.该predict方法还获取样本列表和相应的标签,但这次我们需要返回每个样本的预测:
from sklearn.base import BaseEstimator
class FaceRecognizerModel(BaseEstimator):
def __init__(self):
self.model = cv2.createEigenFaceRecognizer()
def fit(self, X, y):
self.model.train(X,y)
def predict(self, T):
return [self.model.predict(T[i]) for i in range(0, T.shape[0])]
然后,您可以在大量验证方法和指标之间进行选择以测试cv2.FaceRecognizer.您可以在sklearn.cross_validation中找到可用的交叉验证算法:
为了估计cv2.FaceRecognizer我建议使用分层交叉验证的识别率.您可能会问为什么有人需要其他交叉验证方法.想象一下,你想用你的算法进行情感识别.如果您的训练集中包含您测试算法的人的图像,会发生什么？您可能会找到与该人最接近的匹配,但不会找到情感.在这些情况下,您应该执行与主题无关的交叉验证.
使用scikit-learn创建分层k-fold交叉验证迭代器非常简单:
from sklearn import cross_validation as cval
# Then we create a 10-fold cross validation iterator:
cv = cval.StratifiedKFold(y, 10)
我们可以选择各种各样的指标.现在我只想知道模型的精度,所以我们导入了可调用函数sklearn.metrics.precision_score:
from sklearn.metrics import precision_score
现在,我们只需要建立我们的估计和通过estimator,X,y,precision_score和cv以sklearn.cross_validation.cross_val_score,它计算的交叉验证分数我们:
# Now we'll create a classifier, note we wrap it up in the
# FaceRecognizerModel we have defined in this file. This is
# done, so we can use it in the awesome scikit-learn library:
estimator = FaceRecognizerModel()
# And getting the precision_scores is then as easy as writing:
precision_scores = cval.cross_val_score(estimator, X, y, score_func=precision_score, cv=cv)
有大量指标可供选择,请随意选择另一个指标:
所以我们把所有这些放在一个脚本中!
# Author: Philipp Wagner <bytefish@gmx.de>
# Released to public domain under terms of the BSD Simplified license.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of the organization nor the names of its contributors
# may be used to endorse or promote products derived from this software
# without specific prior written permission.
#
# See <http://www.opensource.org/licenses/bsd-license>
import os
import sys
import cv2
import numpy as np
from sklearn import cross_validation as cval
from sklearn.base import BaseEstimator
from sklearn.metrics import precision_score
def read_images(path, sz=None):
"""Reads the images in a given folder, resizes images on the fly if size is given.
Args:
path: Path to a folder with subfolders representing the subjects (persons).
sz: A tuple with the size Resizes
Returns:
A list [X,y]
X: The images, which is a Python list of numpy arrays.
y: The corresponding labels (the unique number of the subject, person) in a Python list.
"""
c = 0
X,y = [], []
for dirname, dirnames, filenames in os.walk(path):
for subdirname in dirnames:
subject_path = os.path.join(dirname, subdirname)
for filename in os.listdir(subject_path):
try:
im = cv2.imread(os.path.join(subject_path, filename), cv2.IMREAD_GRAYSCALE)
# resize to given size (if given)
if (sz is not None):
im = cv2.resize(im, sz)
X.append(np.asarray(im, dtype=np.uint8))
y.append(c)
except IOError, (errno, strerror):
print "I/O error({0}): {1}".format(errno, strerror)
except:
print "Unexpected error:", sys.exc_info()[0]
raise
c = c+1
return [X,y]
class FaceRecognizerModel(BaseEstimator):
def __init__(self):
self.model = cv2.createFisherFaceRecognizer()
def fit(self, X, y):
self.model.train(X,y)
def predict(self, T):
return [self.model.predict(T[i]) for i in range(0, T.shape[0])]
if __name__ == "__main__":
# You'll need at least some images to perform the validation on:
if len(sys.argv) < 2:
print "USAGE: facerec_demo.py </path/to/images> [</path/to/store/images/at>]"
sys.exit()
# Read the images and corresponding labels into X and y.
[X,y] = read_images(sys.argv[1])
# Convert labels to 32bit integers. This is a workaround for 64bit machines,
# because the labels will truncated else. This is fixed in recent OpenCV
# revisions already, I just leave it here for people on older revisions.
#
# Thanks to Leo Dirac for reporting:
y = np.asarray(y, dtype=np.int32)
# Then we create a 10-fold cross validation iterator:
cv = cval.StratifiedKFold(y, 10)
# Now we'll create a classifier, note we wrap it up in the
# FaceRecognizerModel we have defined in this file. This is
# done, so we can use it in the awesome scikit-learn library:
estimator = FaceRecognizerModel()
# And getting the precision_scores is then as easy as writing:
precision_scores = cval.cross_val_score(estimator, X, y, score_func=precision_score, cv=cv)
# Let's print them:
print precision_scores
上面的脚本将打印出Fisherfaces方法的精度分数.您只需要使用image文件夹调用脚本:
philipp@mango:~/src/python$ python validation.py /home/philipp/facerec/data/at
Precision Scores:
[ 1. 0.85 0.925 0.9625 1. 0.9625
0.8875 0.93333333 0.9625 0.925 ]
结论是,使用开源项目可以让您的生活变得轻松!示例脚本有很多要增强的功能.您可能想要添加一些日志记录,以查看您所在的折叠.但它是评估您想要的任何指标的开始,只需阅读scikit-learn教程,了解如何操作并使其适应上述脚本.
我鼓励每个人都使用OpenCV Python和scikit-learn,因为你可以看到这两个伟大项目的接口非常非常简单.