Kuz’min S.A. Multistep algorithms of image segmentation: principles of development and process visualization

Abstract: The article reviews a problem of a video stream frames segmentation in a “bottom-upwards” approach, which can’t be solved in a single processing step due to errors in binarization. For achieving the desired accuracy some additional processing units are needed, the working of each one end with a shift of working receiver operating characteristic (ROC). Intermediate positions of the displaced characteristic is saved in the operating points set up after each shift. The set of operating point makes a path to the area of required accuracy. This it’s only needed to match the sequence of processing units to get the required accuracy. The author developed a classification of approaches to segmentation in the “bottom-upwards” analysis that allows to determine the most effective processing units. A system of video analysis meeting the most recommendations was developed. Each block of segmentation is represented by a family of algorithm allowing to determine object coordinates with required accuracy, including subpixel accuracy. Fot the described system of video analysis the article presents it’s path and criteria for selecting operating points during the setup. The discussed classification approaches to segmentation along with criteria for selecting operating points and visualization method can be used in developing other systems of video analysis.

Keywords: