Multiclass Classification of Camouflage Images Using Combined WLD and LPQ Feature Set Using a ANN Classifier

Locating objects by creating a bounding box and labelling it with the object name is easy but difficult for camouflaged images. The current work focuses on multi-class image classification using a feature extraction process in this context. The algorithm was simulated on a newly defined dataset CAMIW, which includes images with camouflaged people, particularly for war areas. In such environments, due to improper camera calibration, low sunlight, and no proper viewpoints, the images captured from a distance lack in clear view of objects. So, we have tried to extract features using WLD and LPQ feature descriptors and merged the resultant features. The work involves analysis and getting new and improved feature extraction and reduction in the data dimension based on PCA. The feature maps are then sent to a classifier to find out the class of that particular image. The proposed algorithm is compared with three existing feature extractors and shows a good classification accuracy compared to others for camouflaged images.