Face Recognition System for Blur Image Using Backpropagation Neural Networks Approach and Zoning Features Extraction Method

摘要

Applications of face recognition systems have been fairly well-established. However, since Facial Image Recognition System basically depends on the source images as object recognition and methodology, good analysis results from abnormal images containing a blur or noise have still problems. It is therefore, important to develop efficient method to recognize these abnormal images. To analyze the abnormal image, we used introduction of Backpropagation Neural Networks called BNN. Prior to using BNN approach, segmentation as pre-processing and image extraction using zoning method through filtering, grayscaling, thresholding and zoning calles FGTZ were conducted. To confirm the effectiveness of our study, blur imageswere varied from level 1-5 with 10 variations including variations of poses and lighting. The results showed that pre-processing techniques and extraction methods can generate representative facial features whereas the overall of system containing various blur levels and poses can distinguish well between male and female. Further, the system developed has been successfully recognize faces with an average accuracy above of 79% under various blur levels and variation poses.