Directional Lifting Wavelet Transform Algorithm Based On Image Block Detection

The conventional 2-D discrete Wavelet transform (DWT) is carried out as a separable transform by cascading two 1-D transforms in the vertical and horizontal direction. Such a separable transform cannot fully utilize image features. The directional lifting wavelet transform can effectively take advantage of the directionality and anisotropy of images to achieve better image compression ratio. The existing directional lifting wavelet transform algorithm need do the prediction for all image blocks in all candidate directions to decide which direction is the best. These methods have not fully utilized the features of different image blocks, are very time consuming. Aiming at this problem, this paper improve algorithm with the main contents as follows.The directional wavelet transform can well utilize the rich information of the inhomogeneous regions with many edges and veins, while there is little difference between the directional wavelet transform and conventional DWT for homogeneous regions. According to this characteristic, a directional lifting wavelet transform algorithm based on image block detection is proposed in this paper. The proposed algorithm detect all the image blocks of image to decide whether the image block is homogeneous or not, before wavelet transform. For homogeneous image block, the conventional 2-D DWT is used, while inhomogeneous image block is decomposed using directional lifting wavelet transform.For detecting the image blocks, we research on two image block detection algorithms, the instantaneous coefficient of variation algorithm and the structure tensor algorithm. The instantaneous coefficient of variation adopted in the instantaneous coefficient of variation algorithm can prevent fake edges in homogeneous regions from being detected, so it is suitable for the image block detection purpose. The structure matrix adopted in the structure tensor algorithm is useful for the coherence detection of image region. The image region of strong coherence is usual homogeneous. The experimental results have shown that the two methods for detecting image block proposed in this paper can effectively detect the directional information of the image.The image compression experimental results have shown that compared to the existing directional lifting wavelet transform algorithm , the proposed algorithm reduce the time computational complexity of wavelet transform, and maintain the similar image compression performance. The proposed algorithm significantly outperforms conventional 2-D DWT in lossless image compression and lossy image compression.