Search for Small-Size Objects against the Background of a Nonuniform Random Texture

An algorithm of searching for small-sized (about 10 pixels) objects with constant brightness on a nonuniform random background has been developed. The algorithm is based on the search for the conditional maximum of the product of probabilities of certain brightness values for an object and a local background. For the background, these probabilities are estimated from a histogram constructed for an image segment. The algorithm has been implemented for the case when all object brightnesses in the acceptable range are equally probable. The computational experiment was performed on twenty $$1200\times 1200$$ -pixel photographs of clouds with circles of a radius of two pixels. Of these 200 objects, 169 were found, with the number of type II errors being 280.