Research On Hand-Drawn Image Colorization Technologies Based On Optimized Segmentation

Image is an important media in information transmission and exchange,while color is the key factor to decide the expressive force of an image. Soimage colorization techniques have been widely studied in computer visionand digital image processing in recent years. Generally, image colorization isdivided into two main categories，non-interactive and interactive. Most of theexisting colorization algorithms are limited to images with homogenousregions and continuous outlines. However, due to the randomness andcomplexity of lines and patterns in hand-drawn cartoons, colorization turns tobe a practical and challenging problem. This dissertation mainly discusses andstudies interactive colorization techniques for hand-drawn images．In this paper, an interactive image colorization method is proposed forhand-drawn images, which is not only capable of dealing with complicatedpatterns and lines, but also able to ignore some imprecise user scribbles. Toresolve the grayscale discontinuity which results from complexity andrandomness of edge structure in hand-drawn images, firstly, we use anedge-preserving filter and a Laplace edge detector to smooth colordiscontinuities and strengthen outlines. Then stroke energies are calculatedaccording to initial color strokes to overcome imprecise placement. Finally,grayscale and stroke information are modeled into an energy function, and based on Graph Cut theory an optimal segmentation is obtained. Without theimpact of pattern, boundary gap and imprecise scribbles, the method canproduce more correct segmentation results and can be applied to many kindsof hand-drawn images.Many experimental results have proved that this method, with moreprecise segmentation results, obtains high correctness for various kinds ofhand-drawn images.