IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.4, April 2009 292 Manuscript received April 5, 2009 Manuscript revised April 20, 2009 Hybrid Image Thresholding Method using Edge Detection Febriliyan Samopa † and Akira Asano †† , Department of Information Engineering, Graduate School of Engineering, Hiroshima University, Higashi-Hiroshima, Hiroshima, Japan Summary The main disadvantage of traditional global thresholding techniques is that they do not have an ability to exploit information of the characteristics of target images that they threshold. In this paper, we propose a hybrid thresholding method that combines the P-tile method with an edge detector to assist it in the thresholding process. This method successfully generates more accurate object shape extraction than the conventional methods. Key words: Image Thresholding, P-tile, Edge Detection. 1. Introduction In many applications of image processing, pixel values belonging to the object are substantially different from those in its background. Thresholding is one of the simplest and most commonly used technique to separate the foreground from its background [1][2][3]. Thresholding techniques can be categorized into two classes: global thresholding and local (adaptive) thresholding. In the global thresholding, a single threshold value is used in the whole image. In the local thresholding, a threshold value is assigned to each pixel to determine whether it belongs to the foreground or the background pixel using local information around the pixel. Because of the advantage of simple and easy implementation, the global thresholding has been a popular technique in many years. Several successful thresholding methods based on histogram techniques have been proposed, for example, the methods proposed by Kittler and Illingworth [2], Otsu [4], and the P-tile method [5]. Thresholding techniques based on entropy measures [1][6][7][8] and fuzzy approaches [2][9] have also been proposed. The main disadvantage of traditional thresholding techniques is that they do not have an ability to exploit information of the characteristics of the images that they threshold. They treat all images in the same way, regardless of the specific nature of the images. For some situations, this ‘one-fits-all’ approach is sufficient. However, when greater accuracy and more consistent performance are required, more information should be used to assist the thresholding process. This paper proposes a method of utilizing shape information to assist thresholding process. We combine the P-tile global thresholding method with some edge detection methods to retrieve shape information for assistance, and demonstrate its usefulness in various situations. This is a promising approach because it generates more accurate thresholded images than conventional methods especially for applications that need to extract the object shape. 2. P-tile Thresholding Method P-tile is a shorter form of the word “percentile”. The threshold is chosen to be the intensity value where the ratio of the number of pixels whose value is higher than the threshold to the total number of pixels in the image is closest to the given percentile. The P-tile method is one of the earliest thresholding methods based on the gray level histogram [5]. It assumes the objects in an image are brighter than the background, and occupy a fixed percentage of the picture area. This fixed percentage of picture area is also known as P%. The threshold is defined as the gray level that mostly corresponds to mapping at least P% of the gray level into the object. Let n be the maximum gray level value, H(i) be the histogram of image (i = 0 .. n), and P be the object area ratio. The algorithm of the P-tile method is as follows: s ← sum( H(i) ) # total image area # f ← s # initialize all area as object area # for k ←1 to n f ← f – H(k – 1) # remove k–1 from object area # if ( f / t ) ≼ P then stop T ← k # final threshold value # This method is simple and suitable for all sizes of objects. It yields good anti-noise capabilities, however, it is obviously not applicable if the object area ratio is unknown or varies from picture to picture. Unfortunately, we do not usually have such definite information about the object area ratio. This information