The median filtering method and its improved methods is an effective approach to remove random value impulse noise in images. However

most methods have the same shortcomings in finding the optimal threshold and the edges and over smoothed textures structure of images. In the paper

we propose a novel method based on the geometric structure detection to remove random value impulse noise from images. First

the histogram of a noisy image is used to estimate the noise ratio. Next

the two thresholds are adaptively determined from the noise ratio and the histogram of the detail image. Utilizing these two thresholds

all pixels in the detail image are divided into three sets: ‘uncorrupted pixels’

‘undetermined pixels’ and ‘noise pixels’. The set of ‘undetermined pixels’ is often composed of pixels in edges and textures as well as noise pixels. Finally

the geometric structure detection is proposed to distinguish ‘undetermined pixels’. Based on types of each pixel

the result of the median filtering is modified using the detail image. The simulation results show that the proposed method can remove impulse noise while preserve the edge structure of the image. It is superior to the existing methods in performance.