由單張二維圖片與低解析度距離攝影機重建三維人臉模型

在電腦視覺的領域中，關於人臉的研究一直是相當重要的主題，而三維人臉模型較二維圖片包含了更完整的資訊，因此有著更廣泛且更方便的應用。但三維人臉掃描儀器昂貴且難以取得，使得依據二維資訊重建三維人臉成為一項重要的研究，而通常最容易取得的二維資訊便是人臉正面圖片。現今已有技術可由單張二維圖片重建三維人臉模型，其方法為根據一組三維人臉的統計資料，以二維圖片為目標進行合成，但合成結果受限於使用的統計資料，因此並不一定適用所有人種。近來隨著距離攝影機的逐漸普及，人們可使用儀器來量測物體的深度資訊。但高解析度的雷射掃描器一樣昂貴且難以取得，故我們使用易取得的低解析度距離攝影機來擷取臉部深度資訊，並修正其低解析度與雜訊的問題，擷取出人臉大致的深度分佈，並以此修正二維圖片建出的模型。實驗結果顯示，若重建目標的大致特徵與統計資料相似，(例如同為白人者會有類似的臉部特性)，則二維圖片可重建出相當近似的三維模型。若重建目標的大致特徵不與統計資料相似，仍可建出大致輪廓之後，再以由距離攝影機擷取出的深度來修正，使合成結果更接近實際人臉形狀。我們的研究利用可取得的有限資訊來達成最佳的效果，並使用低成本的方法來實現，成為一個可普遍使用的三維人臉模型重建技術。

關鍵字

電腦視覺；三維人臉重建；三維人臉模形；距離攝影機

並列摘要

Nowadays, there are extensive and convenient applications of 3D Face Model. But the 3D face scanners are expensive and not general instruments, and it makes face reconstruction by 2D information an important research topic. There has existed technique for reconstructing 3D face model by single 2D image. According to the 2D image, the 3D model is reconstructed based on the statistics of a set of 3D faces. But the result is constrained by the 3D face set, and therefore may not be suitable for all people. As more and more development of range cameras, the measurement of depth information is feasible. But Laser range cameras with high resolution are still expensive and not general. Therefore we use the general low resolution range camera to measure the depth of faces, and solve the problems of low resolution and noise. We extract the rough characteristics of depth for the refinement of the model reconstructed by 2D image. The experimental result shows that if the rough characteristics of the target face is similar to the statistics, (ex. the white people have similar characteristics of face), then the 3D model can be reconstructed precisely by the 2D image. If the rough characteristics of the target face is not similar to the statistics, we can still build the general geometry of the face, and then refine the model by the depth information extracted by range camera, and make the result much closer to the true shape of the target face.

並列關鍵字

computer vision ； 3D face reconstruction ； 3D face model ； range camera

參考文獻

[1] W.-Y. Lin, K.-C. Wong, N. Boston, and Y. H. Hu, “3D Face Recognition Under Expression Variations Using Similarity Metrics Fusion”, IEEE International Conference on Multimedia & Expo, Accepted, 2007

[3] Kyong I. Chang; Bowyer, K.W.; Flynn, P.J.; “Multiple Nose Region Matching for 3D Face Recognition under Varying Facial Expression”, IEEE Transactions on Pattern Analysis and Machine Intelligence, Volume 28, Issue 10, Oct. 2006, pp. 1695–1700

[5] D. Huang and H. Yan, “NURBS Curve Controlled Modelling for Facial Animation”, Computers & Graphics, 27:373-385, 2003.

[7] K. H. Choi and Jenq-Neng Hwang, “Automatic Creation of a Talking Head from A Video Sequence,” IEEE Trans. on Multimedia, 7(4):628-637, August 2005

[8] Talafová, Renata - Rozinaj, Gregor: “Face Feature Detection for 3D Model of Talking Head with Speech Synthesis”, EC-SIPMCS 2007, 27–30. JUNE 2007

國際替代計量

由單張二維圖片與低解析度距離攝影機重建三維人臉模型

全文下載

主題瀏覽