Non-redundant Tunable-Q Contourlet Transform and Its Application to Image Coding

Abstract We propose a non-redundant tunable-Q contourlet transform for image coding. Existing non-redundant contourlet transforms use multiscale pyramids to iteratively decompose an image into octave frequency bands in the frequency domain. This constant low-Q transform scheme is not suitable for the coding of rich-texture images, in which there are numerous edges and thus rich intermediate- and high- frequency components in the frequency domain. By integrating a non-redundant rational-dilation wavelet transform with the directional filter bank of the contourlet, we construct a new contourlet transform scheme which is not only non-redundant but also provides a really tunable-Q transform scheme by tuning its dilation factor. Using a search algorithm based on information-maximization, one can determine the optimal-dilation-factor (or Q-pattern) for a given dataset. Based on the SPIHT coding scheme, we construct a coding algorithm using the proposed transform. Experimental results show that for texture image coding, our coding method using the optimal Q-pattern is always superior to many existing state-of-art transform coding methods in both visual assessments and quantitative evaluations, including the JPEG and JPEG2000 standards, and the coders based on the wavelet packet transform and wavelet-based contourlet packet transform.