Median Filter For Transition Region Refinement In Image Segmentation

Abstract

Transition region based image segmentation is one of the simple and effective image segmentation methods. This method is capable to segment image contains single or multiple objects. However, this method depends on the background. It may produce a bad segmentation result if the gray level variance is high or the background is textured. So a method to repair the transition region is needed. In this study, a new method to repair the transition region with median filter based on the percentage of the adjacent transitional pixels is proposed. Transition region is extracted from the grayscale image. Transition region refinement is conducted based on the percentage of the adjacent transitional pixels. Then, several morphological operations and the edge linking process are conducted to the transition region. Afterward, region filling is used to get the foreground area. Finally, image of segmentation result is obtained by showing the pixels of grayscale image that are located in the foreground area. The value of misclassification error (ME), false negative rate (FNR), and false positive rate (FPR) of the segmentation result are calculated to measure the proposed method performance. Performance of the proposed method is compared with the other method. The experimental results show that the proposed method has average value of ME, FPR, and FNR: 0.0297, 0.0209, and 0.0828 respectively. It defines that the proposed method has better performance than the other methods. Furthermore, the proposed method works well on the image with a variety of background, especially on image with textured background.