The following is a set of analysis test results, in which novel image streams were analyzed using synthesis modules. In each case, we present the novel image stream side-by-side against the synthesized stream. We also present plots of the analysis parameters extracted by the analysis routines. In general, plots of all the parameter streams are shown, but in some cases, we additionally plot only a few of the important parameter streams, where most of the changes are taking place.

Pose Movements

In this set of experiments, we wanted to test the ability to analyze pose movements. We constructed a synthesis network made of 9 examples that could synthesize left-to-right, and top-to-bottom pose movement. This pose network was then embedded within the analysis routines, and tested on four novel sequences involving various pose movements.

		MPEG	Quicktime	Analysis Parameters
	Left	25K	256K	Pose // All
	Right	24K	208K	Pose // All
	Bottom-Left	28K	292K	Pose // All
	Top-Right	32K	323K	Pose // All

Eyes, Pose, and Mouth Movements

In this set of experiments, we wanted to test the analysis of pose, mouth, and eye movements combined. We created a 14-example synthesis network that could synthesize those three movements, and used them to analyze three test sequences: one involving mouth movement alone, one involving eye movement alone, and one involving a combination of all three types of movements. The result are shown here:

		MPEG	Quicktime	Analysis Parameters
	Mouth	30K	390K	All
	Eyes	48K	557K	All
	Eyes, Pose, and Mouth	38K	347K	All

Mouth Expressions

In this experiment, we wanted to test the analysis of mouth expressions. A 5-example synthesis network was created that could synthesize expressions along two axes: smile, and open-mouth. The network was used to analyze only one test sequence, as shown.

		MPEG	Quicktime	Analysis Parameters
	Mouth	53K	686K	Mouth // All

Affine Movements

Finally, a last set of experiments were made in order to test the extraction of affine movements of the head, such as translation, rotation, and scale changes. The same 5-example expression network constructed for the analysis of mouth expressions, as shown in the previous section, was used to analyze four test sequences involving various affine movements.

		MPEG	Quicktime	Analysis Parameters
	Rotation	42K	437K	Rotation // All
	Translation	25K	259K	Translation // All
	Scale Up	33K	392K	Scale // All
	Scale Down	40K	350K	Scale // All