五官定位與膚色一致性應用於人臉置換系統之研究

Abstract

人臉置換系統在娛樂業是相當受到重視的工作，但現今大部分的工作都是利用手動與特殊的道具輔助，目前存在人臉置換的文獻並沒有很多，在本篇論文中提出一個新的人臉置換系統，目的在於能將人臉置換的工作純粹使用影像處理來完成。 在系統中，利用了膚色分析與形態學(morphology)分割人臉區域，由邊緣偵測(edge detection)與連通量(connectivity)分析，配合唇色分析，找到嘴巴的特徵點，接著利用嘴巴與眼睛的幾何關係與投影函數(projection function)的分析將眼睛定位出來，再由梯度(gradient)與熵(entropy)找出下巴特徵點配合最小平方法(least square method)找到下巴的曲線。根據兩張臉的特徴點比對(feature matching)，我們可以將我們想要置換上的人臉做出適當的形變，並貼在適當的地方。而我們對人臉區域中的膚色部分分離出來，對原始臉與目標臉的膚色區域做直方圖比對(histogram matching)，將原始臉的膚色近似目標臉的膚色，使兩張人臉膚色相近。再借由影像混合(image blending)與平滑化消除接縫。實驗結果顯示，本研究提出的人臉置換系統，當在人臉有存在雙眼的角度中，人臉上沒有太大的光影變化與皮膚質感相差不大的情況下，都有不錯的效果。

Face replacement system plays an important role in the entertainment industries. Recently, however, most of these are assisted by hand and specific tools. Therefore, there are not many papers in this field. In this thesis, we describe a new face replacement system for automatically replacing a face with image processing. In the system, we segment the facial region by skin color analysis and morphology. Feature detection and connectivity analysis are applied to find candidates of facial feature. Then we combine mouth color analysis to obtaining features of the mouth. The eyes are located by geometric relation with mouth and projection function. We obtain features of the chin by information of gradient and entropy. Then we use the least square method to construct the chin line. We can wrap and locate the target face by feature matching. We extract skin color pixels on face region. We make the skin color of the original face similar to that of the original face by histogram matching. Image blending and smoothing is used to eliminate the seam. As can be seen, the experimental results show that this face replacement system has good performance if both original and target faces are front-facing as long as there are no large variations in illumination and skin tones.