Research Of Algorithm Of Stroke-based Image Colorization

Image colorization refers to the technology of adding colors to a gray-level image or video by computers, it has been widely used in old photos manipulation, old paintings restoration, video processing. In this paper, we focus on the stroke-based colorization method, in the view of some shortages still exist in this filed, we put forward two novel colorization methods which can simply the complexity of the human interaction, accelerate the colors propagating and improve the quality of image colorization results. The methods mainly includes two aspects of research workIn this paper, firstly, the importance of color strokes in the image colorization processing is analyzed and studied. This paper proposed a novel colorization model based on segmentation and local and global consistency, using the theory of over-segmentation and the technology of gray histogram to obtain a semi-automatic color stroked image. Finally, with the introduction of machine learning based on the idea of learning with local and global consistency to construct a color of framework based on local and global consistency of learning, color problem is formulated for an optimization problem, to calculate the remaining colored pixel's color value, completed the whole image colorization, and makes the image color is clear and natural.Finally, this paper studies how to speed up the color diffusion of large scale images. This paper presents a method of manifold preserving image colorization using segmentation. The image is first over-segmented, solve the colorization problem on segmentation points instead of pixels, whose number is much smaller than the number of pixels, so the runtime performance is largely improved. Followed by manifold structure constructed using the locally linear embedding (LLE) based on segmentation points in the lightness channel space, then we can obtain the colors of each segmentation points by mapping this manifold structure from the lightness space to color space. Finally, the colors at each pixel can be computed by linear interpolating from the segmentation points at its neighboring regions.