Abstract: This paper proposes a scene-adaptive temperature-emissivity-texture (TeX) decomposition method for near-surface observation based on the thermal texture theory to address the issues of unclear edges and textures in long-wave thermal multispectral images. This approach overcomes the reliance of the existing TeX-semiglobal decomposition method on calibration with diffuse reflectance standard panels and prior knowledge of material spectral emissivity. The proposed method estimates sky radiation through scene-minimum radiance to eliminate calibration dependence, and improves the temperature-emissivity separation framework to achieve the adaptive generation of emissivity curves via correlation clustering, with a joint evaluation system for texture enhancement and target segmentation established for validation. Experiments demonstrated that under calibration-scarce conditions, the proposed method achieved 85% of the calibrated baseline's comprehensive accuracy, significantly enhancing the engineering practicality of thermal infrared images in complex scenes.