PROJECT TITLE :

MSFD Multi-Scale Segmentation-Based Feature Detection for Wide-Baseline Scene Reconstruction

ABSTRACT:

Conventional detectors, such as SIFT, SURF, FAST, A-KAZE, and MSER, have a difficulty with sparse and non-uniform correspondence distribution in wide-baseline matching. A new segmentation-based feature detector (SFD) is introduced in this study, which delivers an enhanced number of accurate features for wide-baseline matching. Bilateral image decomposition can be used to obtain a large number of scale-invariant features that can be used to rebuild a wide-baseline dataset. An existing segmentation technique such as Watershed, Mean-shift, or simple linear iterative Clustering is applied to all input photos. An intersection of three or more regions' boundaries is where feature points are found. The image function's local maxima are what are known as the feature points that have been detected. By using segmentation rather than global thresholds in order to find features, feature detection can be used to detect features throughout the image. Multi-scale SFD improves the matching performance at varying scales, according to a comprehensive evaluation of SFD. The number of features detected and matched between wide-baseline camera views is increased by as much as a factor of 3-5 compared to SIFT, and feature detection and matching performance are maintained with increasing baseline between views. Comparing SFD to SIFT/MSER/A-KAZE for sparse multi-view wide-baseline reconstruction, an increase in reconstructed points by a factor of 10 may be seen, as can better scene coverage. SFD has a higher number of wide-baseline matches with a lower error rate when compared to the ground truth.