A Review of Color Image Vectorization Techniques

Abstract

Digital image processing of gray scale or color images has become an important research and investigation tool in many areas of science and engineering. As the resolution of output devices has risen over the past 20 decades, the demand of high resolution contents has also increased. However, most images are stored in raster format. This format represents images on the computer screen using a rectangular grid of pixels, arranged in rows and columns and the images have a fixed resolution; once the image has been defined at a specific resolution, it cannot be scaled further. Therefore, the image is degraded when enlarged or scaled down which is a major problem in computer graphics, vision, and imaging. Since images with higher pixel density are enviable in many applications, there is an increasing demand to acquire high resolution raster images from low resolution. Raster images are resolution independent therefore; there is a need to convert raster images to vector. Vector graphics use geometrical primitives to express a raster image. These primitives are more compact, editable, scalable, resolution independent and also smaller in file size. Several vectorization algorithms have been developed for the last three decades. This paper presents a review of three categories of raster to vector algorithms, triangulation, mesh-based and parametric patches and algorithms developed using these techniques by various academic authors. Our findings reveal that important algorithms for converting raster images into vector have been developed. However, some of these algorithms perform better than others. Some deficiencies in the algorithms are caused by many possible outlines that can be obtained from the same bitmap image and the technique used by authors. The techniques are powerful but they also have some limitations when representing pixels. Most of the methods produce acceptable results for non-photorealistic images but some are worse or the same as the problem since they blur the image as well especially in photorealistic images. Further research should focus on how to use techniques discussed in this paper to develop a raster to vector algorithm that can convert photorealistic images to vector without compromising image quality.