For video images with large-scale moving foreground objects and rotation jitters, a new fast digital image stabilization algorithm was proposed. According to the poor robustness of the existing circular-projection algorithm to estimate video's rotation jitter with large angles, several modifications for existing circular-projection algorithm was performed to improve the accuracy in the projection correlation matching.As the jitter videos contain larger or much more moving foreground objects, a novel Adaptive Block-selected Matching Algorithm (ABSMA) based on image bit-plane pyramid was proposed to determine the optimal block selection mode according to statistical characteristics of block matching results and to effectively reduce moving foreground objects on global motion estimation. Finally, an adaptive compound filter using the frame grouping and mean filter with a limiting magnitude was introduced, which could not only correct the inter-frame motion estimation error by limiting filter, but also could determine a filtering window to achieve better smoothness of motion by an adaptive frame grouping. Experimental results show that modified circular-projection algorithm has significant effect on improving the accuracy and robustness of image stabilization and the ABSMA can accurately estimate motion parameters of jitter videos with larger or more moving foreground objects. The proposed algorithm is also suitable for the jitter videos without rich gray features and its mean Peak Signal-to-noise Ratio(PSNR) can reach 31.08 dB.