Wavelet Coding With Region

A novel wavelet transform coding method based on the classiication of diierent wavelet amplitude regions is proposed. By using a low-complexity but eecient prediction model, we can distinguish diierent wavelet amplitude regions at upper subbands. Therefore, adaptive quantizers can be applied to diierent regions to achieve better reconstruction quality. The the main advantage of this algorithm is the localized and modular structure which meets the requirement of VLSI implementation. As a result, the employment of our coding scheme makes real-time image processing achievable. In addition, the optimal bit allocation is also discussed to achieve minimum quantization error. Wavel et transform coding has attracted intensive interests in recent years. The principle of wavelet transform coding is to decompose an image into multiresolution subimages. Each subimage corresponds to a diierent frequency band by using multiple scale wavelet bases which results in exible time and frequency resolution. Adaptive quantization can then be used which preserve the statistical property of each upper band subimage to achieve good visual reconstruction quality. This technique is free from blocking artifacts, mosquito noise which exist in block transform coding schemes like JPEG when high compression ratio is required. So far there are lots of algorithms proposed for wavelet transform codingg1,2,3,4].They can be divided into two generations. In the rst generationn1,2], the wavelet-based coding were designed to exploit the decomposition of image into statistical independent subimages. The coding gain was obtained by optimizing the quantizer with eecient bit allocation. However, the inter-band dependencies among subbands were not examined. The new generation of coding schemes focused on the investigation of cross-correlation among subbands to improve the coding eeciency. A well known method for this generation is the embedded zerotree wavelet algorithm (EZW))3]. This approach is based on the prediction of the absence of signiicant information across scales (resolutions) using zerotrees. The major disadvantage of this algorithm is the computational complexity which make it diicult for the real-time processing. Therefore, the investigation of a coding scheme, which can employ the dependencies among subbands eeciently and ooer algorithmic simplicity for real-time processing, has become an urgent issue. In this paper, we exploit the dependencies among subbands by examining the maximum directional variation in each small block of baseband image and the maximum wavelet amplitude in the corresponding block of upper band images. A low-complexity linear prediction model is then applied to predict the maximum wavelet amplitude in each block of upper band …