Synthesising the new video sequence

Once a trajectory has been synthesised in the appearance parameter space, we can generate a new video sequence. Each point on the trajectory corresponds to one frame in the video sequence. For each point, we generate an image of the face by using the appearance parameter ${\bf c}$ in equation 3.10 and use the result ${\bf b}$ to reconstruct the shape (equation 3.7) and the shape-free texture (equation 3.8 page ). The shape-free texture is then warped to the computed shape.

Examples of generated video sequences can be seen on the accompanying CD-ROM. The files:

• examples/V1/V1_wor.m1v,
• examples/V2/V2_wor.m1v,
• examples/V3/V3_wor.m1v

have been generated without using the linear residual model while the files:

• examples/V1/V1_wr.m1v,
• examples/V2/V2_wr.m1v,
• examples/V3/V3_wr.m1v

have been generated using the linear residual model. All those files have been generated using the greedy algorithm.