Towards Efficient for Learning Model Image Retrieval

Abstract

Image mining is widely concerned in processing geo-tagged landmark images of alphanumeric and real-time satellites. Useful information loss in feature extracting process may results in inappropriate image categorization. Reserving useful information is highly challenging and critical in feature extraction and reduction. This research work intends to utilize the hybrid features such as Local Binary Pattern (LBP), colour moments and statistical features for enhancing the categorization accuracy. Then, the k-means classification technique is used to determine the class labels used for model training. In order to mitigate overfitting and to increase the overall classification precision, the Component Reduced Naive Bayesian (CRNB) model is proposed. Also, the physical landmarks of the geo-tagged images are located by using the Hybrid Feature Extraction based Naive Bayesian (HFE-NB) approach. During experiments, two different datasets have been used to test the proposed model, and some other existing models are considered to compare the results. The results stated that the proposed method significantly improves the precision, recall and accuracy of image retrieval. When compared to the existing techniques, it provides the best results by using the texture and colour features with increased sensitivity and specificity such as 3.36% and 0.1 % respectively.