A new method is presented to optimize atmospheric correction for hyperspectral thermal imaging for object detection and recognition. The spectral similarity scale (SSS) is utilized to quantify the "goodness of fit'' for spectral emissivity output from atmospheric correction. The in-scene atmospheric correction (ISAC) method is enhanced with the SSS metric and applied to a MODTRAN 4 perturbed SEBASS sensor scene. Outputs of this process include the spectral emissivity for each pixel of the test scene image consisting of input spectral radiance. Atmospheric perturbations for long slant ranges are found to shift the required reference wavelength significantly from the commonly fixed reference either to the long or short side, dependending on the atmospheric parameters. This shift significantly impacts proper scaling of the corrected image data. Initial evaluation of the new method utilizes measured emissivity spectra of nonblackbody pixels for the ground truth reference to compare with the solved ISAC emissivity spectra. A scene-based receiver operating curve (ROC) metric is applied for minimized values of the SSS to demonstrate levels of improved object recognition for different range values of atmospheric path transmittance and radiance. We also determine future work, which offers the capability to significantly improve the fundamental method for identifying in-scene blackbody pixels, which are required for the ISAC methodology. Finally, the emissivity- based SSS comparison matrix demonstrates the capability to relate the differences in the ROC results to differences in the magnitude of the target spectral similarity.