Evaluation of Multi-Directional SMF Using Image Segmentation and Edge Detection for Large-scale Image

Abstract

Our multi-directional switching median filter is one of several types of filters which detect noisy pixels in an image and denoise them. A threshold setting is necessary with the optimal value depending on the quantity of each image's edge. We focus on this point and propose an improved method that divides an image into segments and, for each one, sets a threshold proportional to the quantity of the edge. Generally, SIDBA images are used as test images. Compared with current large-scale images, test images are smaller in the number of pixels and their subjects are outdated. Therefore, we evaluated the denoising performance of our filter using more realistic large-scale image and this paper reports our results.