The first step of detecting changes between at least two temporal images is geometrically registering two images. After the registration had been performed, then the differences of the images are simply found by subtraction techniques. On the other side, registration task is performed in two steps. At first corresponding features from individual datasets are found with a search strategy based on image matching. Secondly, the datasets are geometrically aligned by using the found corresponding point sets. In order to register images there are various registration techniques. One of the effective matching technique is adaptive least squares matching (ALSM) technique, which is based on the minimization of the radiometric differences of the source and target image patches. The minimization of the radiometric differences, namely the errors between source and target patches are performed with the least squares (LS) adjustment technique. ALSM technique assumes that the source image patch is error free while the target is assumed to be erroneous. However, the source patch should also involve errors. When the errors of the source patch are ignored, these ignored errors reside as uncertainty in the model and thus this situation theoretically must affect the estimation results negatively. When TLS is used, both source and target patches are all assumed erroneous and thus the errors of the data matrix are also taken into account so that it removes uncertainties. We give the first results and discussions of our work on TLS matching.