[Paper Review] Ear Recognition With Score-Level Fusion Based On CMC In Long-Wave Infrared Spectrum

Ümit Kaçar, Mürvet Kırcı|arXiv (Cornell University)|Jan 27, 2018

Biometric Identification and Security10 references3 citations

TL;DR

This paper proposes a score-level fusion approach for human ear recognition using long-wave infrared (LWIR) imagery, leveraging a novel 2430-sample LWIR ear database (MIDAS) and combining multiple feature extraction and dimensionality reduction techniques. The method achieves a 97.71% rank-1 recognition rate and introduces the concept of 'perfect rank'—the rank at which recognition reaches 100%—critical for forensic applications such as corpse identification.

ABSTRACT

Only a few studies have been reported regarding human ear recognition in long wave infrared band. Thus, we have created ear database based on long wave infrared band. We have called that the database is long wave infrared band MIDAS consisting of 2430 records of 81 subjects. Thermal band provides seamless operation both night and day, robust against spoofing with understanding live ear and invariant to illumination conditions for human ear recognition. We have proposed to use different algorithms to reveal the distinctive features. Then, we have reduced the number of dimensions using subspace methods. Finally, the dimension of data is reduced in accordance with the classifier methods. After this, the decision is determined by the best sores or combining some of the best scores with matching fusion. The results have showed that the fusion technique was successful. We have reached 97.71% for rank-1 with 567 test probes. Furthermore, we have defined the perfect rank which is rank number when recognition rate reaches 100% in cumulative matching curve. This evaluation is important for especially forensics, for example corpse identification, criminal investigation etc.

Motivation & Objective

• To address the scarcity of research on human ear recognition in the long-wave infrared (LWIR) spectrum.
• To develop a robust, illumination-invariant ear recognition system operational in both day and night conditions.
• To create a new, large-scale LWIR ear database (MIDAS) with 2430 records from 81 subjects for benchmarking.
• To evaluate score-level fusion techniques for improving recognition accuracy in LWIR-based ear recognition.
• To introduce and analyze the concept of 'perfect rank'—the rank at which recognition rate reaches 100%—for forensic relevance.

Proposed method

• The authors collected 2430 LWIR ear images from 81 subjects to form the MIDAS database, enabling robust evaluation under varying conditions.
• Multiple feature extraction algorithms were applied to extract distinctive ear features from LWIR images.
• Subspace methods were used to reduce feature dimensionality while preserving discriminative information.
• Classifier-specific dimensionality reduction was applied to align feature spaces with the requirements of individual classifiers.
• Score-level fusion was employed by combining the best-performing scores from multiple classifiers to improve recognition performance.
• The cumulative matching curve (CMC) was used to evaluate performance, including the novel metric of 'perfect rank'.

Experimental results

Research questions

• RQ1 Can score-level fusion improve recognition accuracy in long-wave infrared ear recognition?
• RQ2 How does the proposed method perform on a newly created large-scale LWIR ear database (MIDAS)?
• RQ3 At what rank does the recognition rate reach 100%—and why is this 'perfect rank' significant for forensic applications?
• RQ4 How robust is the system to variations in illumination and time of day?
• RQ5 To what extent do different feature extraction and dimensionality reduction techniques contribute to the final recognition performance?

Key findings

• The proposed score-level fusion method achieved a rank-1 recognition rate of 97.71% on 567 test probes from the MIDAS database.
• The system demonstrated high robustness to illumination changes and operational effectiveness in both day and night conditions due to the use of long-wave infrared imaging.
• The concept of 'perfect rank' was introduced and analyzed, providing a new forensic-relevant metric for evaluating recognition systems.
• The MIDAS database, consisting of 2430 records from 81 subjects, was successfully created and used to validate the proposed approach.
• The fusion of multiple classifiers' scores significantly outperformed individual classifier performance, confirming the effectiveness of the fusion strategy.
• The results indicate strong potential for LWIR ear recognition in forensic applications such as corpse identification and criminal investigations.