A new fast method for digital image stabilization

Abstract

There are many methods for digital image stabilization. In this paper, a fast algorithm for the digital image stabilization is proposed that use background small area for digital image stabilization. Background areas are some parts of image that have not motion object and they are suitable for global motion estimation. by using background area most of the computations for calculation of global motion are reduced and we can used local motion vector in background area directly as global motion vector. All background area are not good for motion estimation. We must use only the non uniform background area because usually motion estimation in uniform area produces incorrect result. In our proposed method 16 block in all corners (4 block for each corner) are selected then by proposed method one of these blocks is selected as best block(non uniform block that belong to the background) then local motion vector(LMV) is calculated only for this block, and it is used as global motion vector(GMV) and it is used to stabilize that frame. in next frame at first we test conditions of previous location block for new frame and if it provide our conditions we use that block for new frames so many computations have eliminated otherwise new block location must be search that provide our conditions. until selected block has conditions of our algorithm. It is used, When it has not the conditions we search new block that provides our conditions. In proposed method full search algorithm is used for local motion estimation[1], but instead of searching points in a block for full search we use, partial distortion elimination (PDE)[2] method that was used to terminate the improper candidate blocks and reduce computation for block matching in full search method.