The trinary image tree: A flexible edge-aware multiscale structure

Traditional multiscale structure filters decompose image into two channels as low and high frequency. However, it is difficult to deal with some edge related operations because edge is still implied in high-frequency band. In this paper, we propose a new edge-aware filter which provide an explicit edge layer by obtaining the difference between edge-preserving bilateral filter and Gaussian filter. To separate edge and noise from image further, a flexible purpose-oriented trinary tree with recursively decomposition was built. Corresponding to multiscale analysis, we provide flexible synthesis method to address our trinary tree applied in many different problems. Our trinary tree filtering method offers better performance in contrast to bilateral filter and Gaussian filter. This has been proved by our edge-preserving image denoising experiments.