Wild Image Retrieval with HAAR Features and Hybrid DBSCAN Clustering For 3D Cultural Artefact Landmarks Reconstruction

In this digital age large amounts of information, images and videos can be found in the web repositories which accumulate this information. These repositories include personal, historic, cultural, and business event images. Image mining is a limited field in research where most techniques look at processing images instead of mining. Very limited tools are found for mining these images, specifically 3D (Three Dimensional) images. Open source image datasets are not structured making it difficult for query based retrievals. Techniques extracting visual features from these datasets result in low precision values as images lack proper descriptions or numerous samples exist for the same image or images are in 3D. This work proposes an extraction scheme for retrieving cultural artefact based on voxel descriptors. Image anomalies are eliminated with a new clustering technique and the 3D images are used for reconstructing cultural artefact objects. Corresponding cultural 3D images are grouped for a 3D reconstruction engine’s optimized performance. Spatial clustering techniques based on density like PVDBSCAN (Particle Varied Density Based Spatial Clustering of Applications with Noise) eliminate image outliers. Hence, PVDBSCAN is selected in this work for its capability to handle a variety of outliers. Clustering based on Information theory is also used in this work to identify cultural object’s image views which are then reconstructed using 3D motions. The proposed scheme is benchmarked with DBSCAN (Density-Based Spatial Clustering of Applications with Noise) to prove the proposed scheme’s efficiency. Evaluation on a dataset of about 31,000 cultural heritage images being retrieved from internet collections with many outliers indicate the robustness and cost effectiveness of the proposed method towards a reliable and just-in-time 3D reconstruction than existing state-of-the-art techniques.