Revision history [back]

HaarDetectObjects have many params, sometimes it fails to find face or detects not face.

1. "The function scans the image several times at different scales" how many scales used?
2. What images been used for training cascades in OpenCV?How rect size was selected?Only a face without a chin and hair?
3. How to tune optimal params for my own test image set? For example I mark test image set by hand with rectangles, and then somehow test performance comparing detected rects.

HaarDetectObjects have many params, sometimes it fails to find face or detects not face.

1. "The function scans the image several times at different scales" how many scales used?
2. What images been used for training cascades in OpenCV?How rect size was selected?Only a face without a chin and hair?
3. How to tune optimal params for my own test image set? For example I mark test image set by hand with rectangles, and then somehow test performance comparing detected rects.

HaarDetectObjects have many params, sometimes it fails to find face or detects not face.

1. "The function scans the image several times at different scales" how many scales used?
2. What images been used for training cascades in OpenCV?How rect size was selected?Only a face without a chin and hair?
3. How to tune optimal params for my own test image set? For example I mark test image set by hand with rectangles, and then somehow test performance comparing detected rects.

Update: About training phase.

Training illumination varying faces in one detector resulted in pretty poor. The generated detector became sensitive to illumination rather than robust to illumination. This detector does not detect non-illuminated normal frontal faces. This makes sense because normal frontal faces did not exist in training sets so many. Training multi-view faces in one time resulted in the same thing. thing.

So how to build generalized detector? Big face database is not enough, if we have large variation in data? What parameters should I choose if I don't want to miss any of faces but don't care about false positives?

HaarDetectObjects have many params, sometimes it fails to find face or detects not face.

1. "The function scans the image several times at different scales" how many scales used?
2. What images been used for training cascades in OpenCV?How rect size was selected?Only a face without a chin and hair?
3. How to tune optimal params for my own test image set? For example I mark test image set by hand with rectangles, and then somehow test performance comparing detected rects.

Update: About training phase.phase. http://note.sonots.com/SciSoftware/haartraining.html

Training illumination varying faces in one detector resulted in pretty poor. The generated detector became sensitive to illumination rather than robust to illumination. This detector does not detect non-illuminated normal frontal faces. This makes sense because normal frontal faces did not exist in training sets so many. Training multi-view faces in one time resulted in the same thing.

So how to build generalized detector? Big face database is not enough, if we have large variation in data? What parameters should I choose if I don't want to miss any of faces but don't care about false positives?