This paper proposes a two-stage global motion estimation method and a hybrid coding algorithm for the model-based coding. In the first stage, global motion is estimated by the feature-based algorithm. The estimated result is further refined by the gradient method. The two-stage estimation algorithm is performed hierarchically to remove the influence of the local facial expression. In the estimation process, we also utilize the steerable pyramid to improve accuracy. The facial expression region is coded by the clip-and-paste method, and is predicted by a classified prototype coding technique. The prototype coding technique can greatly improve the coding efficiency for the facial expression. The areas, which are difficult to be described by generic models, are encoded by the proposed hierarchical motion segmentation algorithm. We segment the image to different moving areas which can be modeled by affine models in a split-and-merge manner. The segmented results with the estimated affine models can give good prediction for the unmodeled regions. Computer simulation results show that the proposed complete model-based coding scheme gives very good performance in terms of peak signal-to-noise ratio and compression ratio.

Conventional standard video coding schemes based on independent coding of nonoverlapping image blocks produce undesirable effects (blocking effect and color bleeding, etc.) at low bit rate. A hybrid coding scheme evolving from the standard MPEG codec that combines lapped orthogonal transform (LOT) and overlapped block motion compensation and quantization approaches based on human visual system sensitivity is proposed to remove the artifacts. The LOT is applied to compress the resulting prediction errors of the overlapped motion compensation. The LOT coefficients are quantized by a dynamic intra/inter scene adaptive quantizer that is designed according to the human visual system and scene analysis to optimize the decoded image quality. This technique involves automatically deriving the most favorable quantizer matrix from the motion and scene analysis of image sequences and scaling the resulting matrix further by considering the buffer status. If a scene change is detected from the motion information, the proposed adaptive quantization can also help solve the difficulty of the buffer controller and maintain the picture quality as constant as possible. Performance of the proposed hybrid coding scheme with scene adaptive quantizer is demonstrated by computer simulations using standard test sequences and compared to that of the conventional MPEG-2 coding schemes. Subjective and objective tests show that image quality is improved considerably and the required bit rate is much less than that of the conventional method. Particularly, the image quality can be kept almost consistent, and the undesirable artifacts are reduced considerably.

To represent an image of high perceptual quality with the lowest possible bit rate, an effective image compression algorithm should not only remove the redundancy due to statistical correlation but also the perceptually insignificant components from image signals. In this paper, a perceptually tuned subband image coding scheme is presented, where a just-noticeable distortion (JND) or minimally noticeable distortion (MND) profile is employed to quantify the perceptual redundancy. The JND profile provides each signal being coded with a visibility threshold of distortion, below which reconstruction errors are rendered imperceptible. Based on a perceptual model that incorporates the threshold sensitivities due to background luminance and texture masking effect, the JND profile is estimated from analyzing local properties of image signals. According to the sensitivity of human visual perception to spatial frequencies, the full-band JND/MND profile is decomposed into component JND/MND profiles of different frequency subbands. With these component profiles, perceptually insignificant signals in each subband can be screened out, and significant signals can be properly encoded to meet the visibility threshold. A new quantitative fidelity measure, termed as peak signal-to-perceptible-noise ratio (PSPNR), is proposed to assess the quality of the compressed image by taking the perceptible part of the distortion into account. Simulation results show that near-transparent image coding can be achieved at less than 0.4 b/pixel. As compared to the ISO-JPEG standard, the proposed algorithm can remove more perceptual redundancy from the original image, and the visual quality of the reconstructed image is much more acceptable at low bit rates.