Abstract : Most facial recognition computer systems, including two-dimensional and three-dimensional systems, follow a basic algorithm. The algorithm consists of analyzing nodal points. Nodal points, in this case, are specific pixels on the face highlighting various facial features. These points combined are called a faceprint. Once the computer creates a faceprint for a captured face, it will try and match it to a face in the database. However, there is no guarantee that computers will be correct.
During this research, the objective was to determine the probability that a novel GUI-based software program can match a subject’s captured facial photograph to the same subject’s photograph in a database, and to determine a facial recognition system’s accuracy. Two images of a voluntary sample of subjects were acquired. One set of images’ Sum Of Weighted Ratios (SOWR) value was saved to the program’s database and the second set acted as the captured images. The SOWR values were determined with internodal distances, ratios and weighted ratios. To decrease bias, a simulation was performed with a sample size of 10% of the population. An in-depth analysis of the average, standard deviation and matched pairs T-Test was performed to determine the significance of the difference in SOWR values. (Ideally, the difference of the two SOWR values of each subject should equal 0). Once statistical significance existed for a match, the expected value, or probability, of finding a match using the GUI-based software program with a margin of error was calculated to be 45%.
|
DOI:
10.17148/IJARCCE.2022.11522
