Characterization of empirical discrepancy evaluation measures

Automatic generation of consensus ground truth for the comparison of edge detection techniques

“…There are three reasons for choosing these algorithms: (1) they represent different approaches to the color edge detection problem, (2) which are based on enhancement and thresholding processes and (3) many of them have been used in previous studies [3,25,27,28,35]. In addition, as a reference point, the Sobel operator was applied to the average and monochromatic image I ¼ 1 3 ðR þ G þ BÞ.…”

“…The characterization of the image processing algorithms is essential to analyse newly proposed techniques and to improve the performance of real world applications of computer vision research [1][2][3][4][5][6][7][8].…”

“…The empirical discrepancy measurements are the most suitable for characterization as they are the only ones which are guided by the expected results [3,7]. Their main difficulty is the need to obtain a reliable reference or ground truth, because there is no finite set of artificial or real images which is complex enough to represent all possible real situations or scenes.…”

Radiometric calibration of vision cameras and intensity uncertainty estimation

“…4(b) for an example), while the other set only imaged planes, and, therefore, implied less difficulty for detecting the correct edges for a strategy based on LOG zero-crossings. Besides, up to four empirical discrepancy evaluation measures were used to find the optimum values: the often-cited Pratt's figure of merit (FOM) [14] with (1) aZ1/9 and (2) aZ1; (3) the discrepancy percentage (D) [15,16]; and (4) the Baddeley measure [17], which was found the best evaluation technique in a recent survey [18] involving, among others, Pratt's FOM. In this way, the result of the experiment did not depend on the optimization strategy employed.…”

Application of Gaussian-Hermite Moments in License

“…function's discontinuity (Fernandez-Garcia, & Medina-Carnicer 2004). Our method can have potential applications in video retrieval, and in other related areas of video information processing.…”