ANALYSIS ON DIRECTIONAL FILTER BANK FOR PRESERVING TEXTURE IMAGE

Abstract

Directional filterbank such as contourlet become popular due its ability to capture two dimensional curves efficiently. In this work, transform with directional filterbank is tested to preserve texture information in texture images. Our initial hypothesis is that, transform with directional filter bank will be able to preserve texture information efficiently when compare to wavelet. However this hypothesis is limited if the texture information contained in the images consist of directional information. Textures in this work are referring to repetition pattern in an image. The directional filterbank selected for testing are Non-uniform Directional filterbank (NuDFB) and Contourlet. The texture images are applied with both transforms and the non-linear approximation method is used to capture the significant coefficients of the transformed pixels. The implementations are then compared with wavelet transform as a benchmark. From the experimental result it can be seen that wavelet implementation still managed to outperform directional filter bank transform at higher bit rate. In low bit rate however some significant improvement can be seen from the Peak Signal to noise ratio (PSNR) especially to texture images that contains directional information. In term of Structural Similarity it supports the results of Peak Signal to noise ration with a rare different values. As a conclusion, the directional filterbank transform able to preserve texture information in texture images especially the directional information but limited to a very low bit rate data.