摘要
現今數位相機的普及化，人們利用它取得照片變得更容易，而照片是保存寶貴回憶的重要資產。由於，在拍照過程的當下可能因為姿勢的偏移導至事後的遺憾。因此，本論文提出基於連續拍照之下，產生一套具有自動臉部儀態最佳化影像處理系統，主要流程分為(1)基於照相機技術進行臉部偵測定位、(2)最佳臉部儀態偵測、以及(3)合成與精細化處理。首先，利用Viloa和 Jones所提出的方式將影像中的人臉進行定位動作。其次，透過Projection方法進行臉部遮罩內的特徵點定位並且產生最佳臉部儀態機制去判定出此臉是正臉或側臉。最後，再使用合成和精細化處理，產生最佳的影像品質。經實驗結果顯示，本論文所提出的方法具有對臉部特徵定位和儀態判斷的能力，並且透過自動化臉部儀態處理將連續影像合成為最佳臉部儀態呈現。對於實驗測試資料方面，整體系統架構是實現於本實驗拍攝的連續性團體照之下，而對於臉部特徵定位與儀態判斷是基於本實驗和IMM的資料庫呈現數據結果。其中，從IMM臉部資料庫取得108張測試影像，包含36個不同人物，各有正臉、左側和右側的臉部姿勢。
並列摘要
Nowadays, digital pictures captured are more and more easy and fast. The photos are important asset which can recall more things. However, these pictures may have the bad image quality due to the head posture moved which may cause by human captured or others factors. In order to solve this problem of face posture after capturing, we propose an approach to achieve face posture optimization in the successive pictures based on computational photography techniques This approach involves the steps of face detection based on computational photography techniques, face posture optimization estimation, and synthesis and refined images. Firstly, we adopt a well-known method that is Viola and Jones (V-J) detector to proceed face location. Secondly, we employ the projection method to locate face features within face mask, and then face posture can be judged whether it is frontal face or not based on the posture estimate mechanism. Finally, image can represent an optimum face posture and quality by synthesis process and refined process. Experimental results verify that this approach has the abilities of face feature localization and posture decision and presents the best face posture in the group pictures by means of the proposed automatic optimization system. The estimation of system performance is divided two parts. (1) For posture estimation, it is presented the correction rate used the IMM database which consists of 36 different peoples and 108 facial test images. (2) For posture optimization, the refined group pictures are presented in this thesis used own database which is the successive group pictures.