Automatic Face thinning and automatic eye enlargement algorithms

Currently, apps with various image effects on mobile phones are prevalent, such as camare360, meitu xiuxiu, and powercalm.The effect implementation of the first version of the image effect software was basically based on the color coloring technology, and there were a fewRelatively goodAlgorithm.

However, the most technical content is the rainbow soft perfect365. In its first release, perfect365 implements facial features.Based on this, automatic eye enlargement, automatic face thinning, and other operations related to face beautification are implemented. However, due to poor selection of parameters,In its first version, the effects of these operations are not natural, especially eye enlargement. Six months later, the new version optimized the parameter selection,The effect is much better than the previous one.

Other software is also not weak, and these features are added in subsequent releases. But they all experienced the same experience as perfect365.Process. Originally, I thought that to achieve automatic face thinning, we needed to accurately locate the face contour and used to build an ASM model. But when I ran the ASM model, I found it was quite time-consuming and gave up! My experiments later showed that automatic face thinning is not necessary for exact positioning.Face contour.

First, of course, face detection and precise human eye positioning are required. After positioning, the local amplification and local distortion algorithms of the image are required.Class algorithms are collectively referred to as morphing. Their earliest references can be traced back to a doctoral thesis in 1993: interactive image warping.This article also providesCodeMy implementation is based on this paper.

If you are interested in the Implementation Details and Parameter Optimization Solutions, please contact me further.

Test image Demo:

Original images and after face detection and human eye positioning:

Eye enlargement:

Eye enlargement + face thinning

Another test image:

 

 

 

The test image is from the network. If copyright is involved, please let us know!