Determining the boundaries of grayscale image

Abstract

A grayscale image contains information about an object of interest. Part of the image, free from the object of interest, belongs to background. Known methods of recognition, in particular, statistical methods involve processing the entire area of image. If the background changes unpredictably, processing and recognition quality of such image becomes problematic because the boundary, that separates the object of interest from background, is unknown. The task of object boundaries detection in grayscale images, with heterogeneous background and in the presence of noise with the purpose of further use of these results in recognition problems is urgent. Detection of object boundary in human visual system happens imperceptible, at a subconscious level, even with a high level of noise, which is caused by variable resolution mechanism in human visual system. The paper considers the method and the algorithm for detection of objects boundaries in grayscale images that are presented as discrete implementations of an unknown piece-smooth functions. For adequate presentation the image is considered as a set of cellular complexes, which makes it possible to reproduce the boundaries of objects as one-dimensional lines without thickness, which, in turn, allows to accurately display the smallest details of the object shape at a given resolution. Variable resolution is used in image processing, which enables automatic image processing in the presence of noise without first specifying its parameters. The results of object boundaries detection experiments in a grayscale image are presented in comparison to the results obtained using known methods.