Improving the Accuracy of Detecting Texture Areas in Multi-Channel Images with Noise

A method for detecting textural objects on multi-channel images with noise is proposed. The method is based on splitting multi-bit digital images into bit planes and representing the bit planes of two channels by three-dimensional Markov chains with two states. The method has two steps. At the first step, the white Gaussian noise is effectively removed due to a more accurate calculation of the statistical characteristics of the image within the three-dimensional sliding window and the effective use of statistical redundancy of multi-channel images. At the second step, the selection of extended textural fragments with homogeneous statistical characteristics is performed. Transition probabilities for three-dimensional Markov chains are used as texture features. The estimation of textural features is carried out using a binary image of one of the higher, most informative, digits of the digital image, which allows us to reduce computing resources. Modeling results confirming the effectiveness of combined image processing of two spectral channels for the detection of textural objects are presented.