To obtain a quantitative assessment of the state of automatic face recognition, we performed a meta-analysis of performance results of face recognition algorithms in the literature. The analysis was conducted on 24 papers that report identification performance on frontal facial images and used either the FERET or ORL database in their experiments. The 24 papers contained 68 performance scores that included 40 performance scores on novel algorithms, and matching baseline performance scores for 33 of the 40 scores. There are three main conclusions from the analysis. The first conclusion is that the majority of experiments do not adequately model challenging problems and their results have saturated performance levels. The second conclusion is that authors do not adequately document their experiments. Only twelve out of the 24 papers in this study provided complete documentation. The third conclusion is that performance results for novel or experimental algorithms need to be accompanied by baseline algorithm performance scores.
Citation: NIST Interagency/Internal Report (NISTIR) - 6719Report Number:
NIST Pub Series: NIST Interagency/Internal Report (NISTIR)
Pub Type: NIST Pubs
face recognition, meta-analysis