Study On The Vectorization Of Raster Images

The vectorization of raster images including two steps: image division and vector getting. Through image division we get the edge or framework of a target in form of point series. Vector getting fits the point series with curves such as arcs, lines and ellipse arcs etc. These curves are joining end to end. Vector getting is a key step in the process of raster image vectorization and is the most important part of this article. In the following part of this article vectorization is equal to vector getting. A good vectorization algorithm can not only fit the point series with little error, but also can reduce the data in large amount. In theory, the less fitting error and bigger compress rate, the better a vectorization algorithm is. But little fitting error and big compress rate always take a large amount of calculating, such as in some optimum algorithms, to find the best result always need many times of iterative calculating. In practice, certain fitting error and compress rate can satisfy the demand, so we should choose a suitable algorithm to avoid unnecessary calculating.The background of this article is the shoe pattern design and enlarge and constrict system (shoe pattern system in abstract). In this system, photos of shoe patterns are transmitted to computer from digital cameras. After image division, we get the outlines of shoe patterns in form of point series. The amount of points of a shoe pattern is very large and there are many noises among these points, these make us difficult to enlarge or constrict the shoe pattern, so these point series should be turned into vectors before a shoe pattern is operated in computer. Fitting error of vectorization should be less than 2mm(5 pixels in the digital image) in this system, and a certain compress rate should be get in condition of keep the feature of the shoe pattern outline. In this article, we present a recursive vectorization algorithm based on the existing ones. This algorithm first separate the outline with cusps and a new way of getting cusp points is presented. This new cusp getting way has a less amount of calculating compared with general ways. After separating, algebraic curves was used to fit digital curves between every two cusps in recursive. The experiment of this algorithm shows that recursive algorithm can control the fitting error and compressrate very well through ending condition, without unnecessary calculating, so this algorithm can satisfy the demand of the shoe pattern system.