| Summary: | 碩士 === 國立中正大學 === 資訊工程研究所 === 99 === Imaging technology has made the capture and display of digital images ubiquitous. A variety of displays are used to view them, and it entails that resizing technique also be widely applied. Straightforward image resizing operators, such as scaling and cropping, often produce deformation and distortion. Recent retargeting approaches aim to resize images in a content-aware manner. The seam carving technique proposed by Avidan et al. is a popular approach in this field of study. Most retargeting approaches analyze only color information; we analyze both color and depth information captured by a Kinect to maintain the structure and preserve important regions.
In this thesis, we present the content-aware image retargeting algorithm with the help of depth information. We introduce a depth-based importance map, such that deformation of the image is guided by this map. This depth-based importance map is computed automatically using a novel combination of gradient, salience, and depth-based measures. Depth-based measures is a weight map, we analysis depth information and use mean shift procedure to find the local peak as the central of the important object, and apply Gaussian distribution to determine the weight. We apply this depth-based importance map as the
input of discontinuous seam carving algorithm and we compute the seam by dynamic programming method. To sum up, by adding the depth-based importance map in discontinuous seam carving algorithm, it visually behaves better in maintaining the structure and preserving important regions. In our experiments, the proposed method produces much better results than some existing approaches.
|
|---|
