A quantitative measure of diversity D(A, B) between two images, A and B, is desirable for a good number of practical applications, as for instance: • For assessing the effectiveness of image restoration methods. Restored images Bare matched with the unimpaired image A, and low values of D(A, B) characterize the better
Piero Zamperoni, В. В. Старовойтов · 1995
• For comparing the results of two image partitions obtained by means of two segmentation methods. One of these partitions can be the ground-truth segmentation, if known. • For comparing the results of two edge detectors yielding two different edge images. If the edge detection is performed upon a noisy image, in many cases the edges extracted from the noiseless image represent a good ground-truth edge image, which can be used as a template for assessing the performance of the edge detector. The immediate edge detector outputs are grey-scale images; thus, thresholding and cleaning operations, which would be necessary for obtaining binary edge maps, need not to be involved in the matching process. • For measuring the amount of deformation of an image with respect to a reference one. Impairments can be originated by fluctuating performances of the imaging device, while geometrical deformations can be caused by changes of the viewing position. • For assessing the performance of lossy image coding methods, by comparing the original with the decoded image.