針對即時非線性影像縮放之重要物件感知影像重置技術

Important Object-Aware Image Retargeting for Real-Time Nonlinear Image Scaling

傅國剛(Kuo-Kang)

指導教授 ： 黃崇勛

國立中正大學/工學院/電機工程學系/碩士(2014年)

摘要

面板技術的進步，使得顯示器的面板大小更顯多元，使得影像縮放的技術更須重視。影像大小尺寸的調整，以往是用等比例的縮放，像是縮放(expanding)或是剪裁(cropping)，但是容易造成畫面的失真或是變形，近年來，興起基於影像的內容去做縮放，使得影像中的重要物件不會因此變形，並減少整體畫面失真。
本論文所提之方法，使用硬體達成所要之目標，硬體化的實現，使我們的能即時處理，也因此我們需思考適合硬體實現的演算法，並且降低硬體成本，不使用frame buffer，相較於過往找尋重要物件的方法，我們減少找尋重要物體的計算量，且適合硬體化的實現，最後達到重要物體保持比例且保持整張圖的完整性。經由邊緣偵測找尋可能之重要物件，且為了減少運算量，我們把圖片從像素層級轉成區域層級，並使用相連元件標籤，找出我們所要的重要物體，我們保持住重要物體的長寬比例做縮小的動作，其餘區塊使用不同的縮小比例配合重要物體，最後達到目的之影像大小，並且加強重要區塊占全圖的比例，以達到突顯的效果。

關鍵字

重要物件 ； 非線性 ； 影像重置 ； 即時

並列摘要

Advances in image technology, making the size of the display panel are more diverse. Because there are multi-sized display panel, we have to emphasize on scaling techniques. In general, to adjust the size of image, we use proportional scaling such as expanding or cropping. But they may cause image distortion easily. In recent years, the content aware image resizing is popular. It makes the important information in the image not to be distortion and reduces the overall picture distortion.
In our thesis, we propose a way which achieve desired goal by hardware. The hardware implementation make our result for real time. We need to think deeply about our algorithm which need to be suitable for hardware implementation and cost down. Our algorithm don’t need any frame buffer. Compare to past way of finding important object, we reduce the amount of computation and be more suitable for hardware implementation. At last, we can keep proportion on the important object and hold on all image information. We use edge detection to find objects which are possibly important object. To reduce the amount of computation, we convert image from pixel level to region level. And next, we use connected component labeling to classify different region. We want to retain the biggest region which is our important object and let the other region like background. We keep the proportion of the important object and scale down ,and let the other region use different ratio to scale down to match the important object. And last, we can get the image which fit the goal size. We also want to enhance the proportion of the important object to achieve the effect of highlight.

並列關鍵字

real time ； nonlinear scaling ； retargeting ； important

參考文獻

[4] Itti, L.; Koch, C.; Niebur, E., "A model of saliency-based visual attention for rapid scene analysis," Pattern Analysis and Machine Intelligence, IEEE Transactions on , vol.20, no.11, pp.1254,1259, Nov 1998

[5] Ma, Y.-F. and Zhang, H.-J., "Contrast-based image attention analysis by using fuzzy growing," in [ACM Intl. Conf. on Multimedia],374-381 (2003)

[9] Li-Qun Chen, Xing Xie Fan, Wei-Ying Ma, Hong-Jiang Zhang, and He-Qin Zhou. A visual attention model for adapting images on small displays. ACM Multimedia Systems Journal, pages 353-364, 2003

[12] Setlur, V.; Lechner, T.; Nienhaus, M. Gooch, B., "Retargeting Images and Video for Preserving Information Saliency," Computer Graphics and Applications, IEEE , vol.27, no.5, pp.80,88, Sept.-Oct. 2007

[13] M. Rubinstein, A. Shamir, and S. Avidan. "Improved seam carving for video retargeting, " SIGGRAPH, USA, 2008