Filtered Two-Bit Transform for Block Based Motion Estimation

Abstract

One-bit transform (IBT) and its improved versions such as Constrained-lBT (C-lBT), Weighted Constrained (WCIBT) invariably utilize filtered versions of image frames in constructing a bit plane. The proposed algorithm extends the use of filter kernel to conventional Two-bit transform (2BT) technique. The difference in the intensity levels of original frame and its filtered version represents the fluctuation of an image and is used as an input frame. The input frames are subjected to block-by-block two bit transform using the corresponding block mean and standard deviation values to convert them into two-bit representations. The experimental results obtained show that the proposed method improves visual accuracy and the original 2BT ME performance, and also outperforms other low bit-depth based ME methods of lBT, C-IBT, and WC-IBT on a macro block level.