Digital Image Colorization Techniques Based On Clustering And Distance Transformation

Digital image colorization,emerging in recent years,is a computer-aided technology of adding color to monochrome static images or video sequences.So far,it has been widely applied in many fields,such as old black-and-white photos,digital entertainment,education, remote sensing,medical diagnosis,digital preservation and restoration,etc.Therefore,image colorization holds the characteristics of promising application prospects and potential value. Starting with a brief overview of existing colorization techniques,we focus on the study of colorization for static grayscale images.The related work and main contributions are summarized as follows:1.Colorization based on local color-hint propagationUnder the assumption that "the shorter spatial distance and the more similar local gray statistical characteristics the pixel holds to a given color scribble,the more color contributions from the given scribble the pixel gets",this thesis gives a colorization method.This proposed algorithm is based on distance transformation and fuzzy clustering.Given some typical color scribbles on the input grayscale image,the colorization process consists of the following steps: color scribble extraction,distance transformation and spatial weighting estimation,fuzzy clustering,and weighted color blending.By propagating the local color hints from the given scribbles to the whole input image,the final color image can be produced.2.Colorization by means of color transfer①Efficient color transfer based on divisive hierarchical clusteringConsidering the problem of neighbor selection by exhaustive searching and the time-consuming transferring procedure in Welsh's method,we introduce a new colorization algorithm,in which divisive hierarchical clustering is used to build the binary-tree structure for the reference color image luminance feature space.Based on "the nearest neighbor rule", the color component from the referenced color image can be efficiently transferred to the input grayscale image.②Multi-level color transfer and gradient correction based on Gaussian pyramidIn order to improve the visual effect about the colorization result,this thesis introduces a multi-level color transferring strategy.By Gaussian Pyramid,a multi-resolution form for both input image and the reference color image is generated respectively.Then the multi-level color transfer is carried out from low resolution to high resolution in order to produce different resolution colorization results.Finally,according to the assumption of texture consistency,color correction process is supplied.Therefore the output color image with pleasant visual effects is produced.