English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

In this paper, we propose a full-automatic algorithm for extracting elongated object from CT image. Given an input image, we use the SWT(Stroke Width Transform) and MM(Mathematical Morphology) to locate the coarse location of target. Thus leading to over-segmentation. In order to refine the segmentation result, Levelset method is applied to make the contour fit the object. Experimental results show that the proposed algorithm gives precise contour in elongated object segmentation. Introduction Detecting object in CT image is a vital step for a number of security inspection applications, such as explosive identification, weapon recognition. It is often difficult to extract object because of low contrast property and the effect of interferent. Image segmentation is a ill-posed problem and often needs to deal with fine-structure objects that hold the specific information. it is necessary to take advantage of target’s shape prior knowledge for extracting the fine-structure. In security inspection, a lot explosives are thin elongated objects, in medical field, vessels are often long and thin. To extract these elongated objects in complex environment, it is essential to study approaches regarding how to effectively extract the fine-structure of the image[1]. A lot of methods have been proposed for different situations, and most segmentation algorithm can be divided into two main sub-steps: (1) detection and localization, (2) refinement and enhancement. In first step, MM is often used to find the target position. The result in Candeias’ experiments shows that MM is superior to other edge-detection methods in extracting inner edge[2]. Peng eliminated a large chunk of background area by applying top-hat transform[3]. In order to relieve the effect of interferent, Andres Serna used a morphological h-maxima filter to get rid of maxima caused by texture information and noise on the upper part of the objects[4]. Wang calculated the location of nearest contour pixel for each pixel, which is used to generate a more accurate skeleton[5]. Guillaume separated the image and text part from document image with connected filter[6]. After the coarse localization, a more precise method can be applied to previous result for refining the segmentation. Osher proposed Levelset method for handling the topologic problem[7] in Active Contour Model. Although this algorithm is of great characteristics, some drawbacks, like time-consuming problem and troublesome initialization, limit its usage. To solve these problems, Chan proposed a new energy based on statistic property, thus making Levelset more stable for different initialization[8]. Liu added local energy to C-V Model for speeding up the segmentation[9]. re-initialization is a time-consuming step during contour evolving, Li removed this step by adding another energy cost, which reduced the time of iteration greatly[10]. In this paper, we present a new algorithm by combining Stroke Width Transform(SWT) and Levelset method. The main purpose is to extract elongated object in CT image. We first use SWT and MM to find the position of elongated object, and then Levelset method is applied to refine the segmentation result. International Conference on Materials Engineering and Information Technology Applications (MEITA 2015) © 2015. The authors Published by Atlantis Press 713 Previous Work Strokes Width Transform Model. SWT is proposed by Epshtein for locating text position on image. The basic idea of this transform is that every pixel on image owns a stroke with a specific stroke width, text pixels usually have nearly constant stroke width. SWT merges pixels of similar stroke width into larger components. We can find the most likely text are by bottom up integration[11, 12]. Stroke width of each pixel on image is based on the gradient vector of pixel, we search from current pixel along the direction of inverse gradient, and stop immediately when encounter a edge pixel. The distance from the last pixel to start pixel is calculated and stored for pixels we visited before(Fig. 1). (a)a stoke on image (b)search from p to q, width w is calculated (c)mark all pixels with w Fig. 1 principle of SWT For pixels of foreground, the distance of each pixel is updated according to the path passing through it. We can use the minimum path length to represent its stoke width, but the distance of pixels near corner can be updated to a incorrect value, so Epshtein proposed a solution to solve this problem: replace the minimum value with the median stoke width along the searching direction, this method can correct the false distance to some extent. Elongated object segmentation The first step for elongated object segmentation is locating the potential area. We use the modified SWT to find the rough object on image, the result is refined by MM, and Levelset method will be applied to the coarse result for obtaining a final segmentation result. Modified Stroke Width Transform. A distance map created by SWT[12] will be used for pixels merging and we can see that incorrect merging result is generated (Fig. 2(c)) because of some pixels being never updated to a right distance. To overcome this problem, neighbor pixels of searching path are tested. This paper uses the following formula to update a right distance into distance map: threshold dir norm q P norm dot < − | )) ( ), ( ( | (1) Where dot means dot product of two vector, norm means normalizing a vector, P is the position of current pixel, while q is the position of test pixel, dir represents inverse gradient of current pixel. The value of threshold is 0.4 in our experiments. (a)neighbor pixels (b)original image (c)result of SWT (d)result of our method Fig. 2 generate distance map In order to gain a good segmentation result, we create a histogram for distance map, and remove the pixels with too little or too large distance value. The result is showed in Fig. 2(d). According to the SWT and the shape prior of CT image, we can suppose that target contain the maximum pixels with the same stroke width, thus leading us to find the most likely searching area, and the stroke width containing maximum pixels should be the stoke which we are looking for.

Automatic elongated object segmentation in CT image using Stroke Width Transform and Levelset