人臉特徵碼其及應用

Facial Trait Code and its Applications

李秉翰(Ping-Han Lee)

指導教授 ： 洪一平

國立臺灣大學/電機資訊學院/資訊工程學研究所/博士(2010年)

https://doi.org/10.6342/NTU.2010.02308

摘要

我們提出了人臉特徵碼來對人臉進行編碼與分析。藉由大量的人臉資料，我們定義了重要人臉局部特徵以及其主要類型，這些特徵與類型能有效描述人臉。人臉特徵與類型的建立有兩個步驟：第一個步驟使用沒有光線與表情變化的人臉來進行特徵類型的定義與重要人臉特徵的選取，這些重要特徵與類型能有效區隔不同人之人臉; 第二個步驟考慮了有光線與表情變化的人臉來建立人臉編碼器，以增加辨識人臉辨識之強健性。人臉特徵碼的編碼結果有兩種型態：整數碼與機率碼，前者所需記憶體較小，後者提供較高之辨識準確度。人臉特徵碼可以應用於人臉身份辨識、性別辨識、表情辨識，以及人臉合成上。
在身份辨識的應用上，人臉首先被編碼，然後再與資料庫中已知人臉特徵碼進行比對而得知其身份。在我們的實驗中，人臉特徵碼與目前幾個主流的人臉辨識方法相比有較高的辨識準確度。我們也將人臉特徵碼的方法實作並整合至一自動化的人臉辨識系統。
至於表情辨識方面的應用，我們進一步的定義了表情類型。表情類型捕捉了同一個人之人臉因表情不同而產生的變化。有了表情類型，我們根據表情類型建立各種不同表情的機率模型，並據此進行人臉表情辨識。
針對身份辨識應用所選出之主要類型能有效區分不同人，但這些類型不見的能有效區分性別。我們也選出了能夠有效區分性別之性別特徵，並使用性別特徵來進行性別辨識。
人臉特徵碼的解碼程序可視應用而有所不同。除了上述應用於人臉辨識的解碼外，我們也提出了應用於人臉合成的解碼程序，並且展示了人臉特徵碼可用來合成栩栩如生的人臉。

關鍵字

人臉辨識 ； 人臉編碼 ； 性別辨識 ； 表情辨識 ； 人臉合成

並列摘要

We propose Facial Trait Code (FTC) to encode human facial images for facial analysis. Extracted from an exhaustive set of local patches cropped from a large stack of faces, the facial traits and the associated trait patterns can accurately capture the appearance of a given face. The extraction has two phases. The first phase is composed of clustering and boosting upon a training set of faces with neural expression, even illumination, and frontal pose. The second phase focuses on the extraction of the facial trait patterns from the faces with variations in expression, illumination, and poses. With two different metrics for characterizing the facial trait patterns, the FTC can take either hard or probabilistic codewords. The former offers a concise representation to a face; the latter enables codeword matching with superb accuracy. The proposed FTC can be applied to face identification and verification, expression recognition, gender recognition, and face synthesis. For the application to face identification and verification, a face is encoded as a codeword, and this codeword is matched against known codewords. Experiments reveal that the FTC outperforms many face recognition algorithms, including a couple considered as the most potential ones in recent years. For the application of FTC to expression recognition, the external patterns that encode the variation among facial images caused by different facial expressions are identified, and the pattern-to-expression probabilities are learned. The facial traits are best for discrimination of different individuals. Besides facial traits, we also select the gender traits that are effective in discriminating different genders and use them for gender recognition. The FTC decoding can be task-oriented. We propose the decoding scheme for face synthesis and demonstrate it can synthesize random, life-like faces effectively.

並列關鍵字

face recognition ； facial coding ； gender recognition ； facial expression recognition ； face synthesis

參考文獻

[1] Bouchra Abboud and Mo Dang. Bilinear factorisation for facial expression analysis and synthesis. IEE Proceedings on Vision, Image and Signal Processing, 152(3):327 – 333, 3 2005.

[2] Chiraz Ben Abdelkader and Paul Griffin. A local region-based approach to gender classification from face images. In Workshop of IEEE Conference on Computer Vision and Pattern Recognition, pages 52–52, June 2005.

[5] Timo Ahonen, Abdenour Hadid, and Matti Pietikainen. Face description with local binary patterns: Application to face recognition. In IEEE Transactions on Pattern Analysis and Machine Intelligence, pages 2037–2041, 2006.

[8] Ergun Akleman and Jon Reisch. Modeling expressive 3d caricatures. In SIGGRAPH, page 61, New York, NY, USA, 2004. ACM. 138

[9] Keith Anderson and Peter W McOwan. A real-time automated system for the recognition of human facial expressions. 36(1):96 –105, feb. 2006.