Background: In most of the applications we expect that security cameras or surveillance
camera systems should work on an around the clock basis, as also described in various patents. Normal
cameras which produce images in visible spectrum are not effective with the absence of natural
light. Even though some solutions are there for such problems like morning time natural illumination
and night time artificial illumination arrangement for the camera, but these solutions are not practically
advisable. Also these solutions have limitations like shadow in the morning or day time. As a result,
system may fail in capturing and identifying objects in the dark areas of the current environment.
One of the solutions for such problems is usage of Infrared IR cameras instead of visible spectrum
Methods: This article proposes a novel method of representing infrared images by the use of edgelet
features for object recognition application. The proposed technique makes use of interval valued representation
for edgelet features of the infrared images. A scheme of identification of the objects based
on the proposed features extraction and representation model is also designed.
Results: Experiments conducted to show the effectiveness of the proposed method on publically
available IR corpuses viz OSU Thermal Pedestrian Database, Multispectral Image Database and Indoors
and Outdoors datasets. Two sets of experimentations are conducted where each set contain
three different trails. In the first set of experiments, we have used 40% of the database for training and
remaining 60% is used for testing. In the second set of experiments, we have used 60% training and
40% for testing. In each trail we have randomly selected training and testing samples. For the purpose
of evaluation of the results, we have calculated precision, recall and f-measure for each trail. The details
of the experiments are shown in the Table 1.
Conclusion: This article presents a novel method of representing infrared images by the use of edgelet
features for object recognition applications. The proposed technique makes use of interval valued
representation for edgelet features for identification of the objects in the infrared images. A method of
identification of the objects based on the proposed edgelet features and interval valued representation
model is also proposed. Since the features are transformed into interval valued representation, the
proposed model drastically reduces the dimension of the feature space which intern reduces the computational
time for object recognition in the infrared images. The proposed algorithm is critically analyzed
on three publically available corpuses. Further an extensive experimentation is conducted on
publically available datasets.