The median filtering (MF) algorithm is one of the most commonly used techniques for denoising salt-and-pepper (SPN) noise, and among these algorithms, the adaptive frequency median filter (AFMF) stands out. However, AFMF handles frequency median selection in a relatively simplistic way and fails to address the issue of misjudging extreme points as noise during the median filtering process. This paper presents several optimizations and improvements based on this algorithm. First, for the calculation of the frequency median, we introduce a weighted strategy that differentiates between odd and even numbers. This adjustment results in a frequency median that more closely approximates the pixel values of the original image. Second, we improve the noise determination for the center point of the window by redefining the extreme values of the region, expanding them to the maximum range represented by pixel bits. Moreover, we design an adaptive window adjustment mechanism that applies different window sizes depending on the SPN density, which effectively reduces image denoising time. Experimental results verify the effectiveness of the proposed method and discuss the significance of filter-based denoising algorithms.