Kiran Bhat

Marios Savvides

We propose to develop a video coding technique based on image-

morphing techniques.The input to our system is a recorded

videostream (and so our system has an offline encoder). The

encoder has a user interface where the user specifies a few

"feature curves" in the first frame of the sequence. For instance, in

a videostream of a human face, the user draws the feature curves

on trackable features like lips or eyebrows, etc. The encoder

implements active contour techniques like snakes to track these

features curves through all the frames. Then, for every nth periodic

frame (n appx equal to 30), the encoder will calculate the best fit

line through each of the curves (snakes). Using these lines, a

correspondence matrix is calculated between the periodic frames,

which will be used to reconstruct a morphed sequence of frames in

between. The encoder uses MPEG-1 to encode the periodic

frames into a bit stream. The encoder also interlaces the

correspondance matrix in between the bitstream of the periodic

frames.

The decoder receives the bitstream and decodes the periodic

frames and the correspondance matrices. It then morphes between

the periodic frames using the correspondance matrix.

We will compare the performance of our morphing system with the

standard MPEG-1 video codec for compression ratios and image

quality.

MIDTERM GOALS:

---------------------------

1) Implementation of snakes to track features. We will perform

various experiments to obtain the optimal snake constraints.

FINALTERM GOALS:

-------------------------------

1) Implement the correspondance matrix calculation and morphing.

2) Integrate the MPEG1 codec into our system.

3) Compare the performance of our system with an MPEG1 coded

sequence of the original video stream.

(Depending on progress, we will try to implement part 1 of final

term goal in our midterm)