In transforming original image frames into two-bit representations, the typical two-bit transform (2BT) needs to calculate variances of the local blocks. This calculation of variances inevitably involves multiplication operations and renders the computational complexity of typical 2BT somewhat high. A constrained 2BT (C2BT) for low-complexity motion estimation (ME) is proposed. By exploiting advantages of the typical constrained one-bit transform (C1BT) and typical 2BT, the proposed algorithm significantly reduces the computational complexity of transformation of image frames into two-bit representations. Also, a corresponding matching criterion for C2BT is proposed to enhance the ME accuracy. Experimental results show that the proposed algorithm enhances the ME accuracy by 0.34 and 0.22 dB compared with 2BT-based ME and C1BT-based ME, respectively.
In this paper, adaptive fast matching algorithms are proposed based on the sub-block ordering. Proposed two
algorithms for the lossless fast matching only reduce the unnecessary matching complexity of motion estimation in the
video coding. By observing the expectation of the sum of absolute differences between the current sub-block and a
reference sub-block in the pixel domain, we derive an absolute difference model which indicates the relationship
between the sub-block distortion and the intra sub-block complexity. In the same way, we consider the inter complexity
among all sub-block with the absolute difference model. From the absolute difference model, we derive the transform
coefficient model for measuring the sub-block complexity in the transform domain based on the Parseval's theorem.
Experimental results show the proposed fast matching algorithms reduce the unnecessary complexity of the full search
algorithm about 20% on average without any loss of image quality.
Desynchronization attacks have been a particularly serious threat to watermarking applications. Therefore, there has been a concerted effort to solve this problem. It was Martin Kutter who first proposed to use watermark itself as a tool for estimating the affine transform parameters by its periodicity, what is called the self-reference scheme. His brilliant idea has led to many variations. This scheme is resistant to general geometric transform attacks and the 'reference' is more difficult to destroy than templates. However, this scheme has some drawbacks such as relatively low capacity and high computational complexity, especially in calculating the autocorrelation function. In this paper, we propose a robust watermarking scheme particularly resilient to desynchronization attacks by embedding the watermark periodically. We solve these two problems of the self-reference scheme. For the problem of low channel capacity, we propose to use a finite field sequence to partition a given block and M-ary modulation. For the other problem of high computational complexity, we introduce bit-wise computations in calculating the autocorrelation function. Also, we separate the estimation of a translation factor and affine transform factors, respectively, which results in performance improvement.