Citation Context

...ue to the typically large differences in shading between neighboring regions in cartoons, edge information provides a strong hint for meaningful region segmentation. We thus use a Canny edge detector =-=[17]-=- to extract edge information to guide image decomposition. However, directly using edge information to find region boundaries and decorative lines has various challenges: . Especially when processing ...

Citation Context

...mented into regions using a trapped-ball method, controlled by this mask. . To achieve interframe coherence, the frames in the sequence are registered by finding the homography, using the approach in =-=[16]-=-. A static panoramic background image is reconstructed by first initializing it with unchanging areas and refined by adding regions identified as belonging to the background. The moving objects are ex...

Citation Context

...rtoons requires accurate segmentation into relatively few meaningful regions. There is a vast literature on image segmentation. Many sophisticated color-based methods, such as mean-shift segmentation =-=[6]-=-, typically generate an oversegmented result with too many regions with flat shading and meaningless shape when applied to cartoons. Commercial software, such as Adobe Live Trace, CorelTrace, and Auto...

Citation Context

... background before detection of foreground objects and their motions. Background subtraction has been extensively studied for complex scenes which, e.g., have dynamic backgrounds or changing lighting =-=[11]-=-, [12], [13]. Typically, two separate processes are used for background subtraction and foreground region extraction. While generally performing well, they often neglect pixels near boundaries, which ...

Citation Context

...ould be, but in practice, the mask will not form contiguous, “watertight” boundaries, precluding simple flood filling. One possibility is to try to make the mask watertight by performing edge-linking =-=[19]-=- on the mask. Unfortunately, such methods cannot guarantee to close all gaps, and even a single failure may result in two large regions being inappropriately fused. Another approach is to design a lea...

Citation Context

...before detection of foreground objects and their motions. Background subtraction has been extensively studied for complex scenes which, e.g., have dynamic backgrounds or changing lighting [11], [12], =-=[13]-=-. Typically, two separate processes are used for background subtraction and foreground region extraction. While generally performing well, they often neglect pixels near boundaries, which may cause fl...

Citation Context

...sion artifacts and occlusion. Segmentation which is not coherent between frames will cause flickering in the vectorized output. Previous approaches to temporally coherent video segmentation [8], [9], =-=[10]-=- have tried to avoid such problems via optimization. Although such methods may be appropriate for real-world video, they do not produce precise enough results for cartoons, as even minute inappropriat...

Citation Context

...mpression artifacts and occlusion. Segmentation which is not coherent between frames will cause flickering in the vectorized output. Previous approaches to temporally coherent video segmentation [8], =-=[9]-=-, [10] have tried to avoid such problems via optimization. Although such methods may be appropriate for real-world video, they do not produce precise enough results for cartoons, as even minute inappr...

Citation Context

... decorative lines. Sykora et al. also assume that decorative lines are the darkest regions in a cartoon, which again does not always hold. One way to link line points is to use alpha-shape techniques =-=[18]-=-. Though this work has theoretical importance, in practical use, the results are similar to those produced by traditional heuristic edge linking. We use a simple and fast algorithm based on traditiona...

Citation Context

...er compression artifacts and occlusion. Segmentation which is not coherent between frames will cause flickering in the vectorized output. Previous approaches to temporally coherent video segmentation =-=[8]-=-, [9], [10] have tried to avoid such problems via optimization. Although such methods may be appropriate for real-world video, they do not produce precise enough results for cartoons, as even minute i...

Citation Context

...are too weak or regions are too complicated, as further discussed in Section 7. The morphological operations needed to carry out the trapped-ball filling process take OðnÞ time using the algorithm in =-=[22]-=-. Error sorting needed during region growing can be done by using an index heap or Fibonacci heap, so the whole segmentation process has linear time complexity. 5 INTERFRAME COHERENCE To achieve a hig...

Citation Context

...zed output from Ardeco has a more posterized appearance than our result: in particular, the face and shoulder of the woman have more noticeable steps in shading. Price and Barrett [23] and Sun et al. =-=[2]-=- give methods for vectorizing images that contain objects with complex but smoothly changing textures; they use meshes to describe those textures. Their work thus has very different motivation from ou...

Citation Context

...to regions. 3 OVERVIEW Temporal video segmentation, which can segment a whole video into independent video sequences, each with a different (possibly moving) background shot, is a wellstudied problem =-=[14]-=-, [15]. We assume that such segmentation has already been performed. We focus on vectorization of a single sequence, comprising a static background, possibly with camera motion relative to it, and for...

Citation Context

...ity video to be transmitted with limited bandwidth and also has potential applications to cartoon restoration. In recent years, various approaches have been proposed for vectorizing rasterized images =-=[1]-=-, [2], and commercial packages exist, e.g., Stanford VectorMagic. However, these do not take into account the particular features of cartoon animations, and in particular, tend to suffer from oversegm...

Citation Context

... do not take into account the particular features of cartoon animations, and in particular, tend to suffer from oversegmentation. Furthermore, none of them considers temporal coherence. Sykora et al. =-=[3]-=-, [4], [5] made significant advances in vectorizing cartoons, but their approach rely on a particular drawing style in which meaningful regions are enclosed by thick bounding lines, which is not appro...

Citation Context

...he resulting vectorized output from Ardeco has a more posterized appearance than our result: in particular, the face and shoulder of the woman have more noticeable steps in shading. Price and Barrett =-=[23]-=- and Sun et al. [2] give methods for vectorizing images that contain objects with complex but smoothly changing textures; they use meshes to describe those textures. Their work thus has very different...

Citation Context

...ot take into account the particular features of cartoon animations, and in particular, tend to suffer from oversegmentation. Furthermore, none of them considers temporal coherence. Sykora et al. [3], =-=[4]-=-, [5] made significant advances in vectorizing cartoons, but their approach rely on a particular drawing style in which meaningful regions are enclosed by thick bounding lines, which is not appropriat...

Citation Context

...ions. 3 OVERVIEW Temporal video segmentation, which can segment a whole video into independent video sequences, each with a different (possibly moving) background shot, is a wellstudied problem [14], =-=[15]-=-. We assume that such segmentation has already been performed. We focus on vectorization of a single sequence, comprising a static background, possibly with camera motion relative to it, and foregroun...

Citation Context

...ke into account the particular features of cartoon animations, and in particular, tend to suffer from oversegmentation. Furthermore, none of them considers temporal coherence. Sykora et al. [3], [4], =-=[5]-=- made significant advances in vectorizing cartoons, but their approach rely on a particular drawing style in which meaningful regions are enclosed by thick bounding lines, which is not appropriate to ...

Citation Context

...round before detection of foreground objects and their motions. Background subtraction has been extensively studied for complex scenes which, e.g., have dynamic backgrounds or changing lighting [11], =-=[12]-=-, [13]. Typically, two separate processes are used for background subtraction and foreground region extraction. While generally performing well, they often neglect pixels near boundaries, which may ca...