Object detection in grayscale images based on covariance features

Analysis approach for detection of specific objects in noisy grayscale digital images is described. It is based on calculation of specific coefficients revealing covariance properties of overlapping fragments of pre-processed images. The approach is aimed at detection of rather small foreign objects over the background of larger ones. Before the actual analysis, pre-processing of the images based on median filtering is performed in order to separate small foreground objects from larger background ones. The method was developed and approved within the project related to development of X-ray systems for industrial inspection.