Abstract:

 

We developed an interactive facial sketch synthesis system. A new statistical combined model, called Direct Combined Model (DCM), is developed to overcome the weaknesses of traditional statistical models widely applied for learning sketch examples. Traditional ones either are lack of concerning the relation between different facial information or have some error-prone nature in their assumptions. Furthermore, existing systems consider only local information rather than including global information, which is more close to human sense. Our DCM enhances the feature point location process and optimizes the synthesis process by coupling two relative datasets (such as facial image and sketch) into one concatenated eigenspace. By using an estimated transformation obtained from DCM, an unknown vector (e.g. sketch) can be reconstructed from an input vector (e.g. facial image). Robust synthesizing results are achieved by using global view information efficiently. Experimental results show that the accuracy of automatically locating feature points is promising, and the synthesized sketch resulting by our system is more vivid.

 

Demo:

 

Synthesized Facial Sketches :

 

ØDemo1  Mouse Move !!	ØDemo2  Mouse Move !!
In this demo, the facial sketches of clips with eye blinking and wrinkles are also synthesized vividly.	The other example is more emotional.

 

 

 

Experimental Results:

　

Performance of Synthesizing Sketch

　

　

Synthesized facial sketches

 

 

Examples of detail synthesis sketches of facial features

　

　

Synthesized facial sketches under partial occlusion

 

 

Synthesized facial sketches under different lighting conditions

 

 

 

Synthesized facial sketches under different pitch rotations

 

 

 

Synthesized facial sketches under different yaw rotations