ESTIMATION OF THE EFFICIENCY OF CONTOUR DETECTORS ON THE BASIS OF A POINT RECOVERY FLOW

Abstract

Relevance. Currently, much attention is paid to development of efficient systems designed for processing, storing and transmitting large streams of information obtained in the form of images of space-time fields in such areas of human activity as cosmonautics, geology and geophysics, cartography, navigation, defectoscopy and many others. In most of these systems, when implementing detection algorithms, real images are used as reference images. For example, this is typical for isolation and tracing of the boundaries of reflected waves in images of seismic fields obtained by searching for oil and gas fields. Representation of reference images in the form of real images allows you to adjust the algorithm to a specific subject area, but complicates the assessment of the quality of the algorithm used and the conduct of its objective comparison with other algorithms that implement similar procedures. The aim of the research is to investigate the effectiveness of selection of the contour pattern of images by different detection algorithms based on a generalized complex criterion proposed by the authors of this work. This will allow us to approach objectively to selection of the best detection algorithm and the optimal values of its parameters for any subject area. Methods and means of research. When conducting experiments, the methods of stochastic modeling of space-time signals and fields were used. The computational experiment itself was conducted in the environment of the «KIM SP» software complex. Results. The authors have carried out the analysis of methods for evaluating search operators and localization of boundaries and summarized the main types of errors. On the basis of the proposed classification of metrics for the efficiency of contour detectors, a generalized comprehensive criterion for assessing the quality of selection of contour patterns in images is considered, the possible components of a generalized quality indicator and formal approaches to the choice of weighting coefficients of metrics are shown. The paper introduces the example of implementation of a generalized quality indicator, including five metrics; its effectiveness is investigated for three quasi-optimal contouring algorithms («Canny», «Marr» and «ISEF») at various noise levels. The conducted research allowed drawing the conclusions on the objectivity of the obtained results and giving recommendations on the use of contouring algorithms.

Keywords

• contour detectors
• operators of search and localization of boundaries
• space-time signals and fields
• stochastic modeling
• reference images
• performance metrics
• two-dimensional point-like recovery flow