Texture Analysis Based Detail Preserving Smoothing Filter
基于纹理分析的保细节平滑滤波器

Smoothing is an important topic in image processing area. However, smoothing and edge preserving are hard to achieve simultaneously. In this paper, we propose an improved adaptive detail preserving smoothing algorithm using multiscale and multidirectional mask to deal with this issue. Before smoothing, complexity of texture is calculated according to the edge map. We ourselves define a measurement of complexity of texture which is simple and fast for operation. For the area with large value of texture complexity, that is the area with details, "stick" is used for smoothing to preserve these minute features, while for the area with small value, block mask is employed to speed and emphasize smoothing. As to using "stick", eight different directions are considered to make the most suitable mask take effect. To achieve that, we take the standard deviation as the criterion, small standard deviation denoting homogeneous area and suiting for being taken as the smoothing mask. This method, to some extent, is effective for both smoothing and edge preserving, mainly because it selects different mask according to the local feature of the image utilizing texture analysis. Compared with other methods, including traditional Median Filter, MWMF, Nagao's Filter, DMF and PDE, this algorithm is easy and faster to implement in computer and has better performance.