We describe a scatter calibration technique which compensates for the noisy scaling across the planes and the over-scaling bias in the scatter estimation for scans with high random fraction and/or low number of counts, a situation often encountered in dynamic imaging on scanners with a large number of lines-of-response (LOR) such as the High Resolution Research Tomograph (HRRT). This calibration technique is based on the observations that the scatter fraction is relatively constant as a function of time. i.e. the amount of scatter is proportional to the number of true counts within each dynamic frame with the same proportionality constant. In this work, we first demonstrate the bias in the scatter scaling using the Single Scatter Simulation (SSS) for frames with high random fractions and low number of counts using a phantom study acquired with the HRRT, and we then present two new approaches to scatter correction scaling: a global scatter calibration (GSC) and a segmented plane-based scatter calibration (SPSC) technique. The improvement achieved with GSC and SPSC was examined by comparing the global scatter fraction for the aforementioned frames between the scatter estimates obtained with and without the calibration. The scatter fraction and the scatter sinogram for each segment were also compared between GSC and SPSC method. A significant bias in scatter fraction was found for frames which contain a random fraction higher than 40% with a number of counts less than 10M. For example, a scatter fraction of as high as 270% was obtained from a frame with 90% random fraction and 2M counts, whereas the correct scatter fraction was approximately 40%. A much more consistent global scatter fraction of about 40% was obtained by both GSC and SPSC as compared to the conventional method, and a smoother scatter estimation was achieved by applying SPSC. The potential improvement in the voxel time activity curve (TAC) by applying the calibration is also shown for a cylindrical phantom, non-human primate, and human brain studies.