Regions of Interest Detection in Panchromatic Remote Sensing Images Based on Multiscale Feature Fusion

Abstract

A global searching solution was often employed in traditional prior-knowledge-based regions of interest (ROIs) detection methods for processing high-resolution remote sensing images, which results in prohibitively complex computing. To solve this problem, this study proposes a faster and more efficient ROI detection algorithm based on multiscale feature fusion, wherein the input image is processed along two feature channels: intensity and orientation. The multiscale spectrum residuals method is proposed to compute intensity saliency. The interpolating biorthogonal integer wavelet transform (IB-IWT) is used to extract orientation features, and the orientation saliency is obtained with thresholding and filtering. A weighted across-scale fusion method is proposed to combine conspicuity maps at different scales into one map while retaining salient regions at different scales. The experimental results reveal that the new algorithm is computationally efficient and provides more visually accurate detection results.