Quantcast
Channel: Latest Results
Viewing all articles
Browse latest Browse all 1089

A Human Visual Experience-Inspired Similarity Metric for Face Recognition Under Occlusion

$
0
0

Abstract

Background/Introduction

Recognizing an occluded face is a challenging task for face recognition systems. Although many methods for dealing with occlusion have been proposed, it is more attractive to build a robust face recognition system that focuses on non-occluded regions. Such systems automatically ignore occluded parts, which is broadly consistent with the human visual experience.

Methods

Based on this idea, a new similarity metric called the average degree of aggregation of matched pixels (ADAMP) is proposed. The discrimination performance of ADAMP is derived from information about the spatial distribution of matched pixels.

Results

The proposed method is evaluated with extensive experiments. Compared with state-of-the-art methods, our method is very competitive in terms of recognition accuracy and computation time. In particular, recognition rates of 99.5 % in the presence of sunglasses and 96.5 % in the presence of scarves can be achieved on a benchmark dataset.

Conclusions

Although ADAMP is relatively simple and has the same time complexity as the Euclidean distance, it is demonstrated to be very robust against occlusion. Recognition results using ADAMP are very competitive with those given by state-of-the-art methods.


Viewing all articles
Browse latest Browse all 1089

Trending Articles



<script src="https://jsc.adskeeper.com/r/s/rssing.com.1596347.js" async> </script>