Shi-Tomasi corner detector for cattle identification from muzzle print image pattern

Abstract

Since the last decade, researchers have been using image processing and machine learning to automatically analyse and identify patterns from images without any human intervention. Machine learning tools entail image classification and an efficient feature detection algorithm in the field of computer vision systems. There are various state-of-the-art feature detectors and descriptors available for an image recognition task in image processing and computer vision. In this article, the Shi-Tomasi corner detector has been explored for cattle identification. Two feature descriptors namely, Scale Invariant Feature Transform (SIFT) and Speed up Robust Features (SURF) are considered for extracting the features from the digital images. A linear dimensionality reduction algorithm to make computation faster, reduces the size of features in all cases. For classification, three classification techniques namely, Multi-layer Perceptron (MLP), Random Forest, and Decision tree are considered. For the experimental work, a dataset consists of 930 cattle muzzle point image patterns of distinct 4 breeds. A combination of Shi-Tomasi, SURF, and SIFT are implemented to achieve a recognition accuracy of 69.32%, 74.88%, and 79.60% with MLP classifier, Decision Tree classifier, and Random Forest classifier, respectively. The performance of the proposed system is evaluated based on the classifier-wise recognition accuracy, false-positive rate, true positive rate, and area under curve. Finally, the accuracy is improved by applying an ensemble method (adaptive boosting). The proposed system has reported an improved recognition accuracy of 83.35% by applying Bootstrapping Aggregation methodology along with Random Forest classifier and a combination of Shi-Tomasi, SURF, and SIFT features descriptor.