MPIFaceDetector is the Ikaros port of the MPISearch face detector in the Machine Perception Toolbox. The module can be used either to find one face and two eyes or to find maximally 10 faces in an image. Positions and sizes are given as fractions of the image size.

Example XML definition

A simple example

      class = "MPIFaceDetector"
      name = "MPIFaceDetector"

Module Connections


INPUTThe input image


OUTPUTThe image with the faces indicated by rectangles
FACEThe face cut out from the input image
EYE_LEFTThe left eye cut out from the input image
EYE_RIGHTThe left eye cut out from the input image
FACE_POSITIONLocation of one face in the image (0-1)
EYE_LEFT_POSITIONLocation of the left eye in the image (0-1)
EYE_RIGHT_POSITIONLocation of the right eye in the image (0-1)
FACESLocation of all faces in the image (0-1) and their sizes; each row consists of x, y, size_x and size_y for one face


The face cut out and eyes should be scaled to a standard size and the position smoothed over time. It is fairly useless right now.

When many faces are detected, the most central one should be selected for output and eye-analysis.

The MPT code should be separately compiled to a library.

6 compiler warnings should be removed.


Christian Balkenius
Lund Univeristy Cognitive Science



blog comments powered by Disqus