关键词: 图像去噪, 脉冲噪声, 隶属函数, 开关中值滤波

Abstract: A switching median filter based on boundary discriminative noise detection is proposed to detect impulse noise precisely and denoise. All the pixels in the filtering window are sorted and divided into three groups：lower intensity impulse noise, higher intensity impulse noise, and uncorrupted pixels. If the considered pixel falls into uncorrupted pixels, it is viewed as a signal pixel without processed further, otherwise, it will be classified as a probable noise pixel. The fuzzy gradient function is defined by using the difference of the gray value of probable noise pixels and signal pixels which have the nearest gray value to probable noise pixels. The fuzzy membership function which is used to identify the real noise pixels from probable noise pixels is defined on the gradient function. By using this fuzzy membership function as the weight function, a new weighted median filter approach is proposed. Simulation results show the filter is effective to preserve impulse noise.

Key words: image denoising, impulse noise, membership function, switching median filter