Research On Key Technologies Of Image And Video Special Effects

With the improvement of people’s living standard, and the development of computer, network, and various photo and video capturing devices, people can easily obtain a variety of raw pictures and videos. How to process these raw pictures and videos to generate a desired new digital contents becomes a problem that needs to be resolved. In this situation, image and video special effect editing technologies are developed to realize these functions. As key technologies of image and video editing, digital matting and seamless composition receive a wide research, and become a hot research field. After years hard work, the theories of this area become more and more abundant. However, there still are many problems unresolved, such as low accuracy foreground prediction in video matting, lack of mechanism dealing with the sharp illumination changes and lack of preserving the spatial and temporal coherence for both matting and color adjusting, etc.In this paper, we focus on the existing problems of image/video matting and seam-less composition, several novel and useful algorithms are proposed, including:(1) A "opacity propagation" is proposed in this paper. It uses correlation information between consecutive frame to improve the foreground prediction accuracy. In addition, a color recovering algorithm is proposed according to the local smoothness of illumination changes. By combining the color recovering algorithm with the proposed foreground prediction algorithm, the negative impact of the illumination changes is resolved and the robustness of the algorithm is greatly improved.(2) A foreground prediction based video matting algorithm is proposed. It employs the shortest distance based algorithm to generate the trimaps more accurate than previous methods. It also employs the large scale kernel windows to improve the opacity prop-agation algorithm which can ensure that the pixels similar colors in the corresponding windows will be assigned similar opacity values. In this way, the temporal coherence can be preserved.(3) A two-level graph cut based color harmonization algorithm and a propagation based color harmonization algorithm are proposed. The former one includes the region-level graph cut and pixel-level graph cut. It guarantees that the disconnected regions of one object that have similar hues can be harmonized correctly and ensures that a smooth harmonization can be obtained. In the latter one, a pre-harmonization concept is added in our algorithm. It first employ a propagation based method to pre-harmonize the input image, and then the harmonized image can be obtained by adjusting the hue distribution of the pre-harmonized image. This algorithm can not only preserve the color coherence, but also preserve the variety of the hue values of the harmonized image.(4) A interaction based image color transfer algorithm is proposed to tackle the weakness that the color harmonization algorithm cannot deal with the illuminations and saturations of the image. It first employs the user interaction to separate the images into several layers, and then the pixel colors in each image level is adjusted by color transfer. Finally, a fusional color is computed as a weighted sum of all image layers according to the probabilities of pixels belonging to each layer. In this method, the image layer separation and fusion can preserve the spatial coherence of the result as far as possible.Sufficient experiments demonstrate that, the novel algorithms proposed in this paper perform better than the existing algorithms in accuracy and robustness, and can be applied to several aspects related to the image and video editing. According to the research of this paper, a simple image and video editing system is implemented which can achieve the foreground extraction and seamless composition operations. On one hand, this system further demonstrates the effectiveness of our algorithms, and on the other hand, it is also a foreshadowing of our algorithm to the real application.