The discrete wavelet transform (DWT) has several advantages of multiresolution analysis and subband decomposition, which has been successfully used in image processing. However, the shift-variant property derives from the decimation process of the wavelet transform, and it makes the wavelet-domain motion estimation and compensation inefficient. To overcome the shift-variant property, a low-band-shift (LBS) method is proposed and a new motion-estimation and compensation method using the LBS method in the wavelet domain is presented.
In the LBS method, the input image is shifted by one pixel along x-, y-, and diagonal axis. Then, these are decomposed and decimated by the analysis filter bank. On the next level low-band signal, the recursive operation is iterated. After motion estimation and compensation between the current and reference signal generated from the LBS method, the zero-tree and adaptive arithmetic coding are applied to the displacement frame difference (DFD). Concurrently, the motion vectors are coded to optimize the rate-distortion.
The proposed method had a superior performance to the conventional motion estimation methods such as spatial-domain and other wavelet-domain motion estimation methods with respect to peak-noise-to-signal-ratio (PSNR) as well as the subjective quality. The performance of the proposed motion estimation method is compared with the conventional motion estimation methods. The comparison results show that the proposed method has PSNR improvements of 0.5 ~ 1.5 dB from the full-search method in spatial domain. Judging from resulting images obtained from each methods, our system has excellent subjective quality than others.
The LBS method can be a model method for the motion estimation in wavelet domain just like the full-search block matching in spatial domain. In order to further improve the LBS, the band-by-band LBS is also proposed and evaluated. It has the better performance than the LBS from the view point of the rat...