System and method for adaptive filtering of images based on similarity

Abstract

An interrogation region, such as a portion of a patient's body, is imaged as an unfiltered pattern of image elements, to each of which is assigned a numerical value such as brightness. One element is selected either automatically or manually as a reference element, which lies in a substantially homogeneous image portion. For ultrasonic imaging, the homogeneous portion preferably corresponds to a region of speckle noise. A reference histogram of the numerical values is then accumulated for all the elements in a region or window about the reference element. All of the remaining elements are sequentially designated as current elements, for each of which a current histogram is accumulated. Each current histogram is compared with the reference histogram using an error function for each corresponding pair of histogram bins. The error function values for all bins, which may be normalized and weighted, are used to generate a similarity value between the current and reference histograms. The more similar to the reference histogram the current histogram is, the more strongly it is assumed that the current element lies in a homogeneous portion of the image, and the greater it is smoothed. All possible error function values may be prestored to increase computation speed. The degree of smoothing may be adjusted by the user. The user preferably also may adjust a comparison factor, which determines how strongly a difference between a current and a reference bin will contribute to the similarity value.