Image Reconstruction Using WTA-ICA Model in Contourlet Transform Domain

Abstract

Abstract-A new image reconstruction method using the WTA-ICA model in contourlet transform domain is discussed in this paper. WTA-ICA is in fact an sparse ICA algorithm, and is simpler and faster under high dimensional computational requirements. Contourlet transform can offer a flexible multiresolution and directional decomposition for images and embody well image structure. Here, for a given image, the contourlet transform is first used to obtain low and high frequency sub-band images at two layers, where each layer behaves four directions. Then, WTA-ICA model is used in contourlet domain to extract low frequency sub-band image features and each layer's high frequency sub-band image features. Further, considered the fusion technique between each layer's high frequency sub-band features, as well as that between the fused high frequency sub-band image features and the first level's low frequency features, the total image fused features can be obtained. Using this fusion feature, an image can be reconstructed well. The quality of reconstructed images is measured by SNR criterion, and compared with WTA-ICA model, the experimental results shown that our method is indeed efficient in image reconstruction task.

A remote sensing image enhancement method using mean filter and unsharp masking in non-subsampled contourlet transform domain

“…The non-subsampled wavelet transform (NSWT) (Mallat, 1989) solved the lack of shift-invariance by downsampled wavelet transform, but the resulting image are often not well reserve the details of the original image features neither in the downsampled nor in the non-subsampled wavelet transform domain, the fundamental reason for which is that the wavelet analysis is not the most optimal function representation methods in the two-dimensional space and cannot depict geometry information in the image. The contourlet transform (CT) overcame this defect associated with the wavelet transform (Do and Vetterli, 2002a, 2002b, 2005; Li and Zhanli, 2014; Swaminathan et al, 2013), as it has good expression performance for two-dimensional images, and is convenient and fast. However, CT also has the drawback of lacking the shift-invariance as the wavelet transform because of the downsampling process.…”