摘要: This paper deals with Automatic Face Recognition (AFR). The main contribution of this work consists in the evaluation our two previously proposed AFR methods real conditions. At first, we compare and evaluate recognition accuracy on well-controlled face database. Then these results a real-world database comparable size. For such comparison, use sub-set newly created Czech News Agency (CTK) is from photos acquired by CTK creation corpus represents second work. experiments show significant differences controlled data. 100% achieved ORL while only 72.7% best score for Further show, how rate influenced number training images each person size We also demonstrate, that decreases significantly larger propose confidence measure technique as solution to identify filter-out incorrectly recognized faces. further very beneficial under