Face pre-processing error

[inerplat]

Face detection doesn't work for a re-captured picture of an LCD screen called a "preview picture."

[inerplat]

It is assumed that LCD screens are in poor pixel condition compared to normal images. I should try Gaussian filter on the image

[inerplat]

The use of Gaussian filter and bilateral filter resulted in ambiguous. 🤔

The above mentioned "preview image" wasn't detected normally, but the face was detected in images that had not previously been found

[1] Fail to detect face in raw image

[2] face detect in images had not previously been found

[inerplat]

And I found a new package. It uses the CNN model to detect faces slowly but easily

Pypi : face_recognition

This package works very well. However, since pre-processing takes too much time, it would be use to re-apply only the photos that failed to detect faces while using the existing cv2 cascade method.

It is like chicken-and-egg problem to use deep learning for pre-processing to recognize faces using deep learning.😂

[inerplat]

The test was tried in the mentioned.

There was one mistake, but the results were interesting.

First of all, the wrong color model setting for the CNN-classified face led to a blue-faced picture, but this made clear to distinguish visual images that performed Cascade and the images that carried out CNN.

CNN-based facial recognition did not make a false judgment which is classify a face that is not a face. Therefore, I think not to use the cascade code but only the CNN face recognition.

However, doesn't need to completely erase the cascade code. If multiple faces are detected on the training data it can be used to delete the image using the advantage of having fast speed