Accelerating the distribution estimation for the weighted median/mode filters

Various image filters for applications in the area of computer vision require the properties of the local statistics of the input image, which are always defined by the local distribution or histogram. But the huge expense of computing the distribution hampers the popularity of these filters in real-time or interactive-rate systems. In this paper, we present an efficient and practical method to estimate the local weighted distribution for the weighted median/mode filters based on the kernel density estimation with a new separable kernel defined by a weighted combinations of a series of probabilistic generative models. It reduces the large number of filtering operations in previous constant time algorithms [1,2] to a small amount, which is also adaptive to the structure of the input image. The proposed accelerated weighted median/mode filters are effective and efficient for a variety of applications, which have comparable performance against the current state-of-the-art counterparts and cost only a fraction of their execution time.