MODIFIED EVALUATION OF PRETT–YASKORSKY IN THE GENERALIZED QUALITY INDICATOR OF CONTOUR DETECTION ALGORITHMS

Abstract

Relevance. Obtaining quantitative estimates of the efficiency of image analysis algorithms is a very urgent problem, including the problem of evaluating the efficiency of contour detectors. The choice of reliable algorithms for extracting contours on images of wave fields is of particular importance when constructing digital twins of oil fields. So, to study borehole and interwell space, it is necessary to use contouring algorithms that have high noise immunity in difficult seismic and geological conditions. The choice of such algorithms should be based on a comprehensive analysis of the quality of the selected contours, including such elements as displacement, break and blurring of boundaries. In this regard, the method of obtaining estimates of the efficiency of contour detectors based on the calculation of the generalized indicator of the quality of boundary detection proposed in the work of I. Boaventura and A. Gonzaga in 2009 is of great interest. The generalized quality indicator is defined by the authors of the article in the form of a vector norm in Euclidean space. One of the coordinates of the vector is the reciprocal of the Pratt metric, which takes into account the integral value of the displacement of the elements of the detected contour. The rest of the coordinates of the vector are formed on the basis of binary classification. The works of D.V. Dubinin, A.I. Kochegurov and B. Göringer introduce the studies of the efficiency of the quality indicator of I. Boaventura and A. Gonzaga by a computational experiment. In the course of the experiment, the reference images were approximated by a two-dimensional point recovery stream, which made it possible to implement the principles of setting up an experiment according to Fischer (the principle of randomization). The analysis of the obtained experimental data showed a number of shortcomings of the generalized quality indicator of A. Gonzaga and I. Boaventura. In particular, at low signal to noise (s/n) ratios, overestimated values ​​of the Pratt's criterion were obtained, which in its turn led to underestimated estimates of the generalized quality indicator. Indeed, being sensitive to local displacements of the boundary, Pratt's criterion reacts poorly to omissions of elements in broken contours, which leads to unjustifiably high values ​​of the criterion at low s/n ratios. This statement was made by A.V. Yaskorskiy in 1987 and found its proof in the results of a computational experiment. Thus, there is a real need to improve the reliability of the detection quality estimates obtained on the basis of the generalized index of I. Boaventura and A. Gonzaga. For this purpose, it is proposed to introduce into the generalized quality indicator instead of the Pratt's criterion its modified analogue (Modification Pratt's Figure of Merit), proposed and considered in detail by us in earlier works. This work is devoted to the study of the generalized indicator of I. Boaventura and A. Gonzaga with the modified Pratt criterion. Methods and means of research: system analysis, stochastic simulation modeling, digital image processing, data visualization methods, as well as the methods of analysis, based on the gradual formalization of models by activating the intuition of specialists. The computational experiment was carried out in the Delphi environment, and the numerical model was validated using the Mathcad environment. The research results will make it possible to choose an objective criterion for assessing the quality of building contour boundaries when creating a digital twin of an oil field.

Keywords

• stochastic modeling
• image processing
• oilfield digital twin
• performance evaluations
• quality criterion
• pratt's criterion
• binary classification