For safeguarding dry storage facilities, a tomography system based on fast and thermal neutron detection was studied in Korea Institute of Nuclear Nonproliferation and Control. The study was conducted laboratory-scale experiments based on a custom built 1/10th scale model cask, He-4 gas scintillation detector array, and multiple 252Cf sources. A filtered back projection (FBP) was utilized to obtain the cask image via MATLAB. The Ram-Lak filter (ramp filter) was employed in FBP for improved the reconstructed image quality. The Ram-Lak filter is the increasing amplitude filter due to the increasing spatial frequency of the image. In spatial frequency, the frequency of brightness change in the low-frequency region is relatively low, and the frequency of brightness change in the highfrequency region is large. Thus, the high-frequency region in the neutron tomographic image is near the neutron sources and the cask, and the low-frequency region is outside of the cask and/or between the source and cask in the study. In order to apply the ramp filter, a Fourier transform is initially performed on projection data, and image reconstruction is performed with the corrected projection data. In this case, the filter is linearly changed. Therefore, a small filter value is applied at lower spatial frequencies to reduce the projection data, and a large filter value is applied at high spatial frequencies to reduce the projection data. The filter scale is a fraction of frequency amplitude, and the filter value applied to the projection data is determined according to the filter scale. This study was conducted for discussion of the image quality due to the effect of the filter scale used for image reconstruction of a neutron tomography system. The results show that in the experiment with one source, the source location was founded when we used the frequency scale of 0.5 and over. In the double or triple source experiment, the source locations and relative activities were found when we used a filter scale of 0.4 to 0.6. When the filter frequency scale of 0.7 to over, the relative activities are hard to know. It can be found that if the filter value is too large or too small, distortion may occur in the reconstruction results. Therefore, it seems reasonable to set a value between 0.4 and 0.6 as the scaling factor for the neutron tomography system. In the future, additional comparative studies will perform validation of the frequency scaling methods.