Facial Recognition and Image Filtering

In addition to adding filter items to faces, we also created a blur filter that utilized 2-D convolution that we discussed in class. The inspiration behind this blur filter was to recreate the blurry faces used when people talk but do not want their faces recognizable.

 

This function is also space invariant, as it does not matter where in the photo the face is. The function implementation runs very similarly to the filter function, except it simply detects a bounding box over the face, and applies a Gaussian blur over that, and replaces that back into the original headshot.

 

The exact amount of blurring can be altered through the fspecial function, which allows for alteration of the Gaussian blur and how much is done.

 

In addition, too much blurring creates a black outline around the edges of the face, which is undesirable for the result we want. Therefore, a buffer is built in to only copy in the center part of the face, and not this outline.

 

The basic system diagram is here: