Abstract:Remote sensing image registration is one of the crucial steps in remote sensing image processing, where ground control information is essential. Maintenance of control point databases is complex and expensive. Consequently, lightweight feature databases are emerging. Lightweight feature databases need to store stable and reproducible features. In this context, region features exhibit a distinct advantage. In feature registration methods, the reproducibility of regional features is typically stronger than with individual points. A popular feature region matching method is currently the combination of maximally stable extremal regions (MSER) and scale-invariant feature transform (SIFT). However, the direct combining of MSER and SIFT has difficulties primarily due to redundancy and overlap in regions extracted by MSER, as well as the conflict in applying texture descriptors on homogeneous regions. In this research, we first suggest a salient MSER detection approach that combines frequency-tuned salient region detection and effective nonmaximum suppression filtering to get rid of redundant information and enhance the stability and dependability of the feature region; afterward, we describe the feature region using the unique, enhanced segment distance-weighted gradient location-orientation histogram, which aims to comprehensively describe the feature regions by incorporating more information about the gradient at the edges of the regions. In the experimental phase, we validate the proposed method using multiple remote sensing images. The experimental results confirm the superiority of the proposed method and demonstrate the significant potential and advantages of feature region matching in the context of lightweight feature databases and remote sensing image registration.

Multi-scale maximally stable extremal regions for object recognition

Topology Based Affine Invariant Descriptor for MSERs

Image Retrieval Based on Affine Invariant Feature in Object Regions

Improved Maximally Stable Extremal Region detector in color images

Multi-Spectral Image Registration and Evaluation Based on Edge-Enhanced Mser

Robust Local Feature Descriptor for Multisource Remote Sensing Image Registration

Multispectral Remote Sensing Image Registration Based on Maximally Stable Extremal Regions

Research on the Application of MSER in Image Local Feature Detection

Robust Region Feature Extraction with Salient MSER and Segment Distance-Weighted GLOH for Remote Sensing Image Registration

Multi-Scale Analysis Of Color And Texture For Salient Object Detection

Registration of images with affine geometric distortion based on Maximally Stable Extremal Regions and phase congruency

Robust Affine Invariant Feature Extraction for Image Matching

Affine Invariant Feature Extraction Algorithm Based on Multiscale Autoconvolution Combining with Texture Structure Analysis

MsLRR: a unified multiscale low-rank representation for image segmentation.

Multiscale Relativity Algorithm for Shape-Based Image Retrieval

Adaptive Multi-Level Region Merging for Salient Object Detection.

Multi-Scale Semantic Model for Unsupervised Object Segmentation

Image Recognition Based on Multi-Scale Geometric Analysis and Kernel Matching Pursuit

Scale-invariant contour segment context in object detection.

Multiscale shape context and re-ranking for deformable shape retrieval

A Novel Multi-Resolution Segmentation Algorithm for Highresolution Remote Sensing Imagery Based on Minimum Spanning Tree and Minimum Heterogeneity Criterion