Research On Video/Image Editing Technology On The Compressed Domain

Video/Image editing has become a very active research area in recent years. Compared to the linear editing method, non-linear editing offers the flexibility of video editing, with random access and easy project organization. Compression techniques are developed to decrease the amount of stored visual information without much effect on its quality. Compressed domain techniques deal with data directly in the compressed domain. The computational complexity is greatly reduced owing to the less amount of processing information and avoidance of expensive inverse discrete cosine transform computation required to convert values from the frequency domain to the pixel domain. The main purpose of this research is to develop non-linear editing system based on the compressed domain.In this dissertation, research efforts are concentrated on the image morphing, image inpainting, and video completion based on the compressed domain. Several novel algorithms are proposed, and the experimental results show that these algorithms are efficient.Selective visual attention is the mechanism of selecting and gating visual information based on saliency. Attention is guided towards particular locations in the visual world under a combination of competing constrains. People can rapidly direct their gaze towards objects of interest in visual environment by selective visual attention. By visual attention model, the amount of information to be processed decreases because user can only analyse those regions that may be relevant to the problem. Therefore, a novel method for salient region extraction is presented based on visual attention mechanism in the compressed domain.Image morphing is a class of techniques that deal with the metamorphosis of one image into another, which generates a series of smoothly continuous intermediate images between two given images. The dissertation presents two image morphing algorithms in the compressed domain. The first method is based on the feature-based method. Salient blocks and edge blocks as features are detected in the DCT domain first. Then feature correspondences are built manually. At last, a mapping function is constructed on two hierarchies. The second method is for the case that features are not distinct and feature correspondences can't be built easily. A morphing method following the idea of fluid simulation is proposed. Results show that the processing time is greatly reduced and morphing based on the visual attention can generate the results that well fit human visual perception.Image inpainting is an interpolation problem that restores an image with missing or damaged parts. The dissertation presents an image inpainting method based on the compressed domain. DCT coefficients of the non-inpainting blocks are utilized to get block features, and the block features are propagated to the inpainting region iteratively. The algorithm can recover the missing or damaged parts according to its surroundings. With the reduction of computational complexity, the quality of inpainting result in the compressed domain is still close to the quality in the pixel domain.Video completion is a process of completing object's missing information in a video. A video completion method on the compressed domain is proposed. It is composed of three steps:motion field transferring, thin plate spline interpolation and combination. Missing space-time blocks can be completed by placing new motion vectors on them and the resulting video sequence have as much global visual coherence with the video portions outside the hole.