Colorization is a computer-assisted technique of adding colors to monochrome images,movies or TV programs. The technique was firstly introduced in 1970 by Wilson Markle,and it has been widely used in movie-making, medical instruments, image enhancement forspace exploring, and many other industrial and scientific fields. Up to now, however, thecolorization technique remains an active and challenging area in the realm of imageprocessing.According to its application and processing mode, colorization is divided into twocategories, namely pseudo colorization and false colorization. The aim of Pseudo colorizationis to enhance the non-color image for inspection with no demand for meaning of the color. Incontrast, false colorization intends to recover the real color of black&white image and toimitate the actual scene. This dissertation mainly discusses false colorization of monochromeimages and videos. After summarizing and analyzing all the available existing techniques offalse colorization, a few new and applicable colorization techniques are developed in thispaper.The main contributions in the dissertation are as follows:1. Theoretically summarizing of current colorization methods. Colorization is treated asan ill-posed problem. Colorization methods are divided into two kinds according to theirregularization forms: the one on the basis of color transfer and the other one by using colorpropagation. The color propagation method is considered to be more reasonable. Theunderlying mechanism of the method is the local coherence of image colors. The controllingmodes of color propagation are classified into two classes: micro-controlling andmacro-controlling.2. According to the theory of color propagation, Unevenness is proposed as propagatingrestriction, based on which an algorithm "Colorization Using Chrominance Blending byUnevenness" is devised. In order to improve the efficiency, a fast method is contrived. Andthen the definition of Unevenness is ameliorated to reduce the effect of sudden change onobject surface. The experiment results show this method is fast, effective and not sensitive to noise.3. A global segmentation of grey scale image is evolved from image colorization, whichuses Unevenness as the rule to clustering from the seeds specified manually. This methodlikes human vision, and it depends on the understanding of the experimenter for the image.According to image colorization, a suggestion is proposed to color image compression to usecolorization technique to decrease the redundancy of color information.4. A new video colorization methods are proposed. It is based on the true motionestimation and image colorization. Because the results of currently reported motion estimationalgorithms are found to be greatly different from the actual motion. Consequently, the existingalgorithms are not appropriate for color transfer. We introduce a spatial coherence constraintof motion field to improve the true motion estimation. And then pick up the most confidenceestimation to transfer color, which will be used as the color seeds for the next frame toperform the image colorization algorithm. |