AbstractAn improved image enhancement approach via nonsubsampled contourlet transform (NSCT) is proposed in this article. We constructed a geometric image transform by combining nonsubsampled directional filter banks and a nonlinear mapping function. Here, the NSCT of the input image is first decomposed for L-levels and its noise standard deviation is estimated. It is followed by calculating the noise variance and threshold calculation, and computing the magnitude of the corresponding coefficients in all directional subbands. Then, the nonlinear mapping function is used to modify the NSCT coefficients for each directional subband, which keeps the coefficients of strong edges, amplifies the coefficients of weak edges, and zeros the noise coefficients. Finally, the enhanced image is reconstructed from the modified NSCT coefficients. Three experiments are carried out respectively on images from subjective vision quality and objective evaluation measures. The first experiment is the algorithm performed on images. The subsequent experiments are the information entropy and spatial frequency. The experimental results demonstrate that the proposed method can gain better performance in enhancing the low-contrast parts of an image while keeping its clear edges.
Block-Extraction and Haar Transform Based Linear Singularity Representation for Image Enhancement
