內容感知的快速影像縮放方法

Abstract

影像縮放用於調整圖片的尺寸，目的是為了在不同大小的解析度上顯示圖片。隨著智慧型手機、平板電腦等行動裝置的普及化，影像縮放技術也顯得越來越重要。 本篇論文提出一個快速的多運算子影像縮放方法。本方法藉由將各個影像縮放運算子分開在不同的階段使用，來達到快速並有效地結合。為了使縮放結果能夠與原圖所傳達的訊息一致，本方法也保護圖片中重要內容的比例不受形變影響。另外，我們也更近一步地和雲端技術作結合，藉由在雲端預先完成繁重耗時的計算，以應付即時的縮放應用。最後，我們透過實驗結果來驗證本方法在時間上以及效能上的優勢。

Image retargeting is used to fit images into different sizes of resolutions. With mobile devices gaining more and more popularity, content-aware retargeting technique is highly demanded. In this thesis, a fast multi-operator retargeting method is proposed. The proposed method exploit advantages of different operators at different stages, which allows our approach to combine these operators in an efficient way. To conform to the original image while retargeting, proposed method also manages to protect the perspective of salient contents from shape distortion. A cloud-based method is further introduced to perform real-time retargeting by combining the proposed method with cloud technique. Experiment results show that our approach is effective and efficient.