Description: Method for image analysis for the purpose of detecting similar (same) objects under conditions of lack of information about the structure of the objects, their characteristics and information about their presence was proposed. Analysis of existing methods shows, that they are based mostly on visual features from existing etalon image to search for. The sequence of image processing operations including downscaling, detection of main colors, based on k-means clustering modification method, followed by color quantification and binarization, search for similar connected components in sense of the geometric properties was proposed. Accuracy, provided by sets of different geometric features to be used during confirmation of area’s identity was investigated. An experimental simulation was done for Blocks dataset, which was created (under different scaling, background and illumination conditions) and labeled specially to cover usage examples and possible problematic cases. Testing confirmed the effectiveness of proposed method for search of the same objects without etalon image with accuracy of about 82%. Limitations of the method include: the objects should be completely in the field of vision and should be visually separated in a two-dimensional mapping plane. It should also be noted that the method does not process other information except visual, which means that different objects of the same color and form in the image will be identically identical in the image. The proposed method may be used to analyze images (both natural and artificial) of technological lines and processes, related to search for visually similar or same objects, analysis of theirs’ positions and quantity.
Keywords: image analysis, connected components, quantization, k-means, geometric features, similar regions, markup, similar objects