- 學位論文
基於側面投影法之內容感知圖像與影片縮放研究
Projection Profile-based Content Aware Image and Video Retargeting
摘要
本論文提出了一基於側面投影之內容感知縮放的方法，對圖像以及影片進行非均勻式的變形縮放。首先，我們先以能量公式 (energy function) 計算出圖像及影片畫面中的重要及非重要部分，能量公式的計算則利用邊緣偵測 (edge detection) 及視覺顯著性能量圖 (saliency map) 之方法來實現，影片縮放則另外加入了畫面差異能量圖 (Frame difference map) 。接著利用能量圖計算出圖像或影片中畫面的垂直與水平的能量投影特徵來統計重要性的數據。如此，圖像及影片畫面各個部位縮放的比例將會是不同的，達到保護了重要內容的非均勻式變形縮放的目的。而本方法也能有效地降低在 seam carving 中產生的不連續性問題。
並列摘要
In this thesis, a projection profile-based algorithm is proposed for content-aware image and video resizing. First, an energy map which describes the image or the video frame important and unimportant parts are calculated by energy functions. The edge detection, the saliency-based visual attention model, and video frame difference are applied in the energy functions. Then the proposed method uses horizontal and vertical projection profiles of the energy map to gather importance statistics. Thus, different scaling of the image or the video frame content is extracted. By this way, the discontinuity and misalignment that usually occurs in the results of seam carving is reduced. Experiments result show that the proposed method is capable of yielding some acceptable results.