The results of the investigation of the integrated performance evaluation method of edge detection based on the two-dimensional renewal stream

The paper presents the results of investigation of the edge detection quality for three algorithms (“Marr-Hildreth”, “ISEF” and “Canny”), obtained using the integrated method proposed in the work of Boaventura and Gonzaga. The studies were conducted in the environment of the program-algorithm complex of stochastic modelling “KIM SP”. The features of the program-algorithm complex are briefly described along with its integrated block diagram given. And the sequence composition of the morphologies structures of the reference space-time signals (RSTS) and their bitmap images are presented. The methods and approaches of the statistical modelling were used in obtaining and summarizing the results of the numerical experiments and the reference images were approximated with a two-dimensional hi-rise renewal stream. The edge detection performance for three algorithms under study (“Marr-Hildreth”, “ISEF” and “Canny”) has been evaluated under different levels of additive noise. The noise was generated by a random data generator with a near-normal distribution. The signal energy-to-noise-dispersion ratio was taken as a criterion of the signal-to-noise ratio. The analysis of the algorithms efficiency was based on the morphologies of RSTS reference images of “A” and “F” type. The results of the investigation are presented as a set of the overall quality factor estimates, the estimated probability of the correct edge detection and omission, as well as the false alarm probability against the signal-to-noise ratio. The calculation of the estimated probability of the certain components of the overall quality factor is attributed first and foremost to their various importance in such application domains as aerospace, navigation, geology, nondestructive testing and others. Comparison of the results obtained for three edge detection algorithms and two types of morphologies enabled not only to provide an objective performance estimation of using the integrated method for comparison of the quality of the search and edge location algorithms, but also to produce the selection guideline in the specific application of a particular algorithm depending on the noise level and the morphology type behind imaging.