Modeling And Recognition Of 3D Facial Expression

Facial expression recognition is one of the key research topics in computer vision,image processing and pattern recognition.The preliminary research mainly based on 2D image data.Because of the human face is a 3D geometry originally,there must be lost some texture information and morphological features in the projection process from 3D to 2D.In addition,2D images are susceptible to factors such as postures,makeup and lighting.Its robustness is poor and the accuracy of facial expression recognition is limited.In recent years,with the continuous development of the 3D images obtain techniques,the facial expression is gradually turn to 3D images form 2D images.With 3D images,more facial expression features could be extracted.Compared to the facial expression recognition based on 2D,the 3D facial expression recognition has the stronger robustness and the higher accuracy.At first,in this thesis we improved the existing collection equipment of 2D images.Then we used the improved equipment to collect 2D images and try to rebuild the 3D facial expression based on photometric stereo.Then this paper presented two fully automatic methods for 3D facial expression recognition.The contents and creative works of this thesis is as follows:The first,we improved the existing collection equipment of 2D images,and then we used the improved equipment to collect 2D images.The original collection equipment was used to collect the 2D images of the tongue,it can't be used to collect the 2D images of the tongue.In this thesis,contrary to the features of 3D images of facial expression,we improved the original system's camera module,the support of light source module and the support of the object module.Based on the improved equipment,we can collect the 2D images of facial expression successfully.The second,based on photometric stereo,we used the collected 2D images to reconstruction the 3D facial expression.According to the improved collection equipment,we collect the 2D facial expression images that meet the conditions of follow-up reconstruction work.Then the 3D facial expression model was rebuild based on photometric stereo.The third,in this thesis a 3D facial expression recognition method based on multi-feature fusion was proposed.We used the automatic extraction algorithm to extract the 20 key points of facial expression features,and then we selected 23 pairs distance features based on the 20 key points,these distance features can express facial expression characteristics.At the same time,we extracted the LBP features around every key point,the LBP features can also be used to express facial expression characteristics.At last,we used SVM classifier to recognize the 3D facial expression.The experimental results show that the proposed method can better reflect the facial expression characteristics by using the local LBP feature around the key point and the 23-point distance feature.The fourth,another method of 3D facial expression recognition based on sparse representation was proposed.The algorithm improved the LBP feature to CBP feature,it used the global data to describe facial expression feature and it has the better performance.In this thesis we compared the advantages and disadvantages of LBP feature and CBP feature.We also compared the classification effect between spares representation and SVM classifier.At last,in order to further optimize the recognition results,the 3D face was divided evenly and we used the CBP feature and sparse representation to recognize 3D facial expression.The experiment has achieved good classification results and it proved that the regional division has important influence on facial expression recognition.Finally,we summarized the whole article.