An improved windowed adaptive mean filtering denoising algorithm based on the eight-way Sobel operator
Yao Tang, Zhaoyang Chen · 2023
Noise reduction algorithms used in camera ISP need to be simple, fast, memory efficient and suitable for hardware implementation, so currently noise reduction algorithms suitable for ISP applications are still more common with improved versions of classical low-pass filters. This paper proposes an improved mean filtering algorithm based on the adaptive eight-way Sobel operator window to address the problem that traditional algorithms blur the edges of an image while smoothing out the noise. The algorithm firstly extracts the edges of the image by the improved Sobel operator to distinguish the flat and edge regions of the image, and then adaptively changes the filtering window according to the characteristics of different regions of the image. The algorithm can effectively filter out the noise while preserving the edge details of the image, achieving a balance between filtering effect and edge preservation. The experimental results show that the peak signal-to-noise ratio of the improved method is improved by 45.44% on average. Compared with other algorithms, the improved method has both noise removal and edge retention capability, and still has better noise suppression and edge retention capability when the image is contaminated by high intensity noise.