Regression-Based and Isophote-Constrained Cloud Removal for Time-Series Remote Sensing Imagery

Optical remote sensing images often suffer from cloud-induced quality degradation. Most cloud removal methods rely on cloud-free reference images, which are rarely available. This paper introduces a time-series cloud removal method using regression fitting and isophote constraints. Key steps include: (1) establishing a regression model between completion group images and the reference image using superpixel-based color and structural constraints; (2) extracting an isophote network to construct a weight matrix and solve a linear system for target image reconstruction. Tested on Landsat 8 datasets with varying cloud cover and terrain, the method achieves visually accurate results, with Mean SSIM $\geqslant$ 0.9457 and PSNR consistently above 31, demonstrating strong color consistency and reconstruction accuracy.