Recognition mode

In the Recognition mode, a confidence meter is shown next to the preprocessed face, so the user knows how reliable the recognition is. If the confidence level is higher than the unknown threshold, it will draw a green rectangle around the recognized person to show the result easily. The user can add more faces for further training if they click on the Add Person button or one of the existing people, which causes the program to return to the Collection mode.

Now, we have obtained the recognized identity and the similarity with the reconstructed face, as mentioned earlier. To display the confidence meter, we know that the L2 similarity value is generally between 0 and 0.5 for high confidence, and between 0.5 and 1.0 for low confidence, so we can just subtract it from 1.0 to get the confidence level between 0.0 to 1.0.

Then, we just draw a filled rectangle using the confidence level as the ratio, shown as follows:

    int cx = (displayedFrame.cols - faceWidth) / 2; 
    Point ptBottomRight = Point(cx - 5, BORDER + faceHeight); 
    Point ptTopLeft = Point(cx - 15, BORDER); 

    // Draw a gray line showing the threshold for "unknown" people. 
    Point ptThreshold = Point(ptTopLeft.x, ptBottomRight.y - 
      (1.0 - UNKNOWN_PERSON_THRESHOLD) * faceHeight); 
    rectangle(displayedFrame, ptThreshold, Point(ptBottomRight.x, 
    ptThreshold.y), CV_RGB(200,200,200), 1, CV_AA); 

    // Crop the confidence rating between 0 to 1 to fit in the bar. 
    double confidenceRatio = 1.0 - min(max(similarity, 0.0), 1.0); 
    Point ptConfidence = Point(ptTopLeft.x, ptBottomRight.y - 
      confidenceRatio * faceHeight); 

    // Show the light-blue confidence bar. 
    rectangle(displayedFrame, ptConfidence, ptBottomRight, 
      CV_RGB(0,255,255), CV_FILLED, CV_AA); 

    // Show the gray border of the bar. 
    rectangle(displayedFrame, ptTopLeft, ptBottomRight, 
      CV_RGB(200,200,200), 1, CV_AA);

To highlight the recognized person, we draw a green rectangle around their face as follows:

    if (identity >= 0 && identity < 1000) { 
      int y = min(m_gui_faces_top + identity * faceHeight, 
        displayedFrame.rows - faceHeight); 
      Rect rc = Rect(m_gui_faces_left, y, faceWidth, faceHeight); 
      rectangle(displayedFrame, rc, CV_RGB(0,255,0), 3, CV_AA); 
    }

The following partial screenshot shows a typical display when running in Recognition mode, showing the confidence meter next to the preprocessed face at the top center, and highlighting the recognized person in the top right corner: