PENENTUAN KOMBINASI KERNEL TERBAIK MENGGUNAKAN MEDIAN FILTER

Abstract

Many factors such as the use of low-quality lens, inappropriate image acquisition technique, and the uncertainty of environment condition affect digital image acquisition process. As a result, noisy image may be obtainedwhich is difficult to be recognized, processed, and analysed. The most occurred noise is salt and pepper which appears as white and black spots in image. This noise is usually removed or suppressed using median filter. However, improper size of median filter kernel may yield in poor result. This work aims to find the most suitable median filter kernel size combination using Breadth search, Depth Searchand Generate and Testmethods in order to obtain the best median filtering result. In this work, four median filter kernel sizes are used, i.e. 3×3, 5×5, 7×7, and 9×9. Filtering performance using each obtained combination is defined by measuring the Mean Square Error (MSE) of the filtered and unnoisy image (ground truth). Result shows that the best median filter kernel size combination is {3×3}. This combination is obtained using Breadth Search Method with the MSE of 27.0308.