Image registration aligns two images geometrically, which is frequently required in medical, computer vision and remote-sensing field. It is a crucial pre-processing step in change detection or growth monitoring using satellite images. Accuracy of change detection depends on accuracy of image registration. For multi-modal, multi-sensor, multi-spectral satellite images one of the challenges for image registration is varying illumination level according to the sensor characteristics. This challenge is addressed by using Histogram of Oriented Gradient (HOG) along with Speeded-Up Robust Feature (SURF). It is shown that illumination variation gives some incorrect matches with SURF only which degrades image registration. Incorrect matches are reduced by using HOG as descriptor in SURF. Supporting simulation results for satellite images are presented which show the improvement in the correct matching rate.