Face-preprocess-tools

Face-preprocess-tools is a collection of tools for preprocessing face images. Images will go through FaceDetection, FaceAlignment, FaceCropping processes.

BaseLine

FaceDetection -> FaceAlignment -> FaceCropping

FaceDetection

Haar Feature-based Cascade Classifiers

@param:

• imageList.txt: a list file, which contains image file paths.

@output:

• faceDetection.success: a list file, which records successful detection.
• faceDetection.error: a list file, which records error detection.
• boundingbox.list: a list file, which contains bounding box information of successful detection images.

$ cat imageList.txt
/workspace/datasets/CASIA-WebFace/5587543/001.jpg
/workspace/datasets/CASIA-WebFace/5587543/002.jpg
...

$ python FaceDetection/faceDetector.py imageList.txt
FaceDetection Done!

FaceAlignment

Deep Convolutional Network Cascade for Facial Point Detection

Two parts:

1. Landmark Detection
2. Face Alignment

Landmark Detection

@param:

• boundingbox.list: the output file boundingbox.list of FaceDetection process.

@output:

• landmark.list: a list file, which contains five facial points of face images listed in boundingbox.list.

$cat boundingbox.list
/workspace/datasets/CASIA-WebFace/5587543/001.jpg 47 203 46 202
/workspace/datasets/CASIA-WebFace/5587543/002.jpg 39 195 51 207
...

$ python FaceAlignment/detectLandmark.py boundingbox.list
LandmarkDetection Done!

Face Alignment

@param:

• landmark.list: the output file landmark.list of Landmark Detection process.
• PATH_Aligned_image_dir: the destination directory path of aligned output images.

@output:

• faceAlignemnt.list: a list file, which records successful alignment.
• Aligned_image_dir: a directory, which contains the aligned output images listed in faceAlignemnt.list.

$ cat landmark.list
/workspace/datasets/CASIA-WebFace/5587543/001.jpg 98.1535536766 111.477871895 155.723092651 106.953521013 130.264041901 137.168443489 103.697720623 165.793231583 153.124426651 165.626620483
/workspace/datasets/CASIA-WebFace/5587543/002.jpg 80.4625348091 113.501712513 137.962782669 107.797487926 101.989186478 146.925302696 89.6215407372 172.050915527 136.488462639 161.69779644
...

$ python FaceAlignment/alignment.py landmark.list CASIA-WebFace-Aligned
FaceAlignment Done!

FaceCropping

@param:

• faceAlignment.list: the output file faceAlignment.list of Face Alignment process.
• PATH_Cropped_image_dir: the destination directory path of cropped output images.
• dst_width: the width of cropped output images.
• dst_heigh: the height of cropped output images.

@output:

• Cropped_image_dir: a directory, which contains the cropped output images listed in faceAlignemnt.list.

$ cat faceAlignment.list
/workspace/datasets/CASIA-WebFace/5587543/001.jpg
/workspace/datasets/CASIA-WebFace/5587543/002.jpg
...

$ python FaceCropping/facecrop.py faceAlignment.list CASIA-WebFace-Cropped 62 62
FaceCropping Done!

Reference

[1]. zhangjie, luoyetx/deep-landmark, 2016-01-26

[2]. RiweiChen, RiweiChen/FaceTools, 2016-03-11