License Plate Image Binarization Based on Chromatic Information and Fuzzy C-Mean Approach

the threshold value is determined by using the maximum betweencluster variance method; however, this algorithm is less effective when the contrast between the background and the characters is low since the spatial correlation between pixels is not taken into consideration (9, 10). Bernsen’s method calculates gray values for the neighborhood area around each point and dynamically determines the threshold (11). This approach is more adaptive than the Otsu algorithm, but its computation efficiency is low and it is prone to the issue of fracture (9, 12). Moreover, methods proposed include histogram analysis binarization (13), a heuristic binarization algorithm (14), a minimum entropy method (15), and a valley search method based on a histogram (16). Still other methods were proposed by Ye and Lian (17 ) and by Zhou et al. (18). Binarization algorithms can be divided into two types—the global threshold method and the local threshold method—each of which has its own advantages and disadvantages. Technically, the binarization algorithm contains two main parts: (a) separation of the character from the background and determination of the threshold value and (b) determination of the background and character information. The threshold value is difficult to determine since it has a certain ambiguity. When the threshold is too low, irrelevant image objects (such as background) might be extracted; in contrast, relevant image objects might be removed if the threshold is too high. In addition, traditional methods for distinguishing characters and background are based on previous knowledge, and these methods cannot work well if the image is polluted. In other words, traditional methods for global or local image classification lack adaptability to environmental random effects. Cluster algorithms classify a license plate into characters and background. Traditional methods presume that the background includes more pixels than characters in grayscale images (under ideal conditions), but with the detrimental effects from the outside environment, this ideal situation is not always the case, thus leading to the inability to segment characters and recognize them correctly. Color images contain more information than grayscale images (19, 20). When the images are subject to external pollution, noise and the original information cannot be distinguished from grayscale images, but from color images, color information can still be measured. Many studies have been conducted on color models (19). Kim et al. proposed a license plate location algorithm using the hue, lightness, and saturation model (21). Tsai and Lee proposed a document image binarization method using intensity and saturation information (22). Both methods show good performance (21, 22). There are still many challenges in binarization algorithms and their performance is unsatisfactory in a complex road environment. In response to this situation, a novel method is proposed that clusters images by using a fuzzy c-mean (FCM) algorithm and classifies License Plate Image Binarization Based on Chromatic Information and Fuzzy C-Mean Approach