Adaptive scalar and vector median filtering of noisy colour images based on noise estimation

To address the problem of removing impulsive noise with different density from colour images, a new filtering algorithm is proposed based on noise estimation as well as adaptive scalar (SMF) and vector median filter (VMF). Two-level noise estimation scheme is adopted for noise detection, where the first-level estimation is based on maximum and minimum intensity value of each colour channel, and the second-level estimation uses weighted directional operators. For noise restoration, the uncorrupted pixels are remained unchanged, and the corrupted pixels of low-to-medium density are restored by the double weighted VMF, where the term double weighted means that the pixels´ spatial distance and magnitude value are weighted together for the vector ordering in the computation of vector median filtering. In addition, the corrupted pixels of high density are restored by the SMF based on M estimator and the neighbourhood processed pixels. According to the estimated noise density, the proposed SMF and VMF are switched adaptively. The experimental results show that the new algorithm can filter the noise effectively while protecting the image colour, contrast and fine details well for the impulsive noise of different density (even as high as 99%).