Multilayer Transformation of Different Resolution for Colour Image Analysis and Segmentation

Abstract

The segmentation methods of colour images are often developed advancing the grayscale segmentation methods with application of the particular colour model. The short chracteristics and disadvantages of these methods are described shortly, in addition with common segmentation problems arising within their application. The publication presents the image segmentation problem analysis from psichovisual position offering an image transformation based on human visual system property to analyze the image sharply in the centre and pointlessly in the peripheria of the retina. Such transformation resolves the image to hierarchical layers with the different resolution dependentendly from the selected central location in the image. Colour analysis is performed through CIE L*u*v* colour space model. The spatial features for the segmentation are extracted by simulation of the human visual cortex’s V3 function, which detects the lines of particular length and orientation. Such properties were implemented using the linear Gabor filter, then thresholding the obtained results and finding the centres of the reference segments. The application example demonstrates the performance of the transformation and feature extraction algorithm using the L* component’s matrix of the colour model. Ill. 6, bibl. 8 (in Lithuanian; summaries in English, Russian and Lithuanian).