NEWS

WHAT'S ON TV NOW

Recorded   Latest Recorded Program Feeds Click for more recently recorded TV feeds Kayotix's latest Audio Demo Sample CD Now Shipping. Click Here to order your free copy !

Revolutionary New Animation Software - Video PaintBox 10.12.04 (Euronews) close X Running Time Approximately: 2 minutes Playlist Order: [N/A] Source Site: HERE Reality transformed into animation - as if by magic. Just one of the abilities of a new software with an aesthetic sense which is able to recognise important aspects of a photo or film and then apply the desired effects. The revolutionary software Video PaintBox, has been developed at Bath University, in England under the guidance of Peter Hall and John Collomosse. In this chapter we give a brief introduction to the "Video Paintbox", a novel system for generating AR animations from video. This system is developed throughout the remainder of Part III (Chapters 6, 7, and 8); each chapter dealing with one of the three individual subsystems comprising the Video Paintbox. In this introduction we recall our discussion in Chapter 2 which highlighted the limited scope of current research into video driven AR, and use these observations to motivate development of the VideoPaintbox. We give a brief overview of the paintbox’s core subsystems, their principal contributions, and the arguments they make for higher level temporal analysis in AR.5.1 Introducing the Video Paintbox We observed in our literature review that although image-space AR techniques have become increasingly common in the literature, the problem of synthesising artistically rendered animations from video remains sparsely researched. A very small number of algorithms [75, 96, 103] exist for automatically producing such animations, and each of these propose a solution to the very specific problem of creating a painterly effect in video whilst maintaining temporal coherence. We argue that the current literature is limited in its approach to video driven AR animation in a number of respects.• First, all algorithms which address the problem of automated painterly animation process video on a per frame sequential basis, rendering each frame of animation using data from only the current and previous frame. Processing begins by di-rectly applying a standard, static AR algorithm to the first frame of video. Most methods [96, 103] then translate brush strokes from frame to frame using a vec-tor field produced using an optical flow technique [7], which supplies a per pixelinter-frame motion estimate. Although the level of temporal coherence is, of course, determined by the optical flow technique implemented, optical flow does not generally perform well in the presence of occlusion, varying lighting condi-tions, or within flat untextured regions [54]. Unfortunately, a general solution to the problem of creating an error-free motion field between two images is un-known to Computer Vision; the problem can be reposed as the under-constrained general “correspondence problem”1[145]. Estimation errors quickly accumulate and propagate forward through the video sequence, manifesting themselves asan uncontrolled “swimming” in the animation. Exhaustive manual correction is typically required to compensate for these errors and improve temporal coher-ence [61]. A faster, but less accurate, approach to producing painterly video was proposed by Hertzmann and Perlin [75], where an RGB differencing operation is used to “paint over” areas of the frame that are deemed to have moved, i.e. ex-hibit large inter-frame pixel differences. Under this algorithm, lighting changes orfast camera movements cause behaviour to degenerate to an independent paintingof each video frame, reintroducing associated difficulties with swimming

More ...