Morphological Preprocessing for Low-Resolution Face Recognition using Common Space

Abstract

There are many researches on face recognition, but most have not produced satisfactory results on very low-resolution images. This study proposes the use of morphological preprocessing to improve the performance of common space approach for face recognition on low-resolution images. The morphological preprocessing consists of Top-Hat and Bottom-Hat Transformations, which capable of extracting small elements and handling uneven lighting on images. The k-Nearest Neighbor is used to recognize the face by measuring the distance of deep CNN features of low and high-resolution images in the common space. Experiment on the Yale Face dataset shows that the use of Morphological Preprocessing can increase the face recognition accuracy by 14.59%, 1.00%, and 2.50% for low-resolution images with sizes 24x24, 36x35, and 56x56, respectively.