Fusion of spectral wavelet and spatial total-variation methods to reduce the noise of hyperspectral images

Hyperspectral images as a source of information can be used for diverse applications in various fields, including target identification, classification, change detection and anomaly detection in urban and non-urban areas. Noise is an inevitable part of a signal which limits the use of hyperspectral images in some applications. Noise removal is one of the most important pre-processing stages in hyperspectral images. In order to remove the noise in hyperspectral images, the data needs to be preprocessed to reduce noise impact on the images. The process and analysis of hyperspectral images is rather complicated because of the high dimensionality of hyperspectral images compared to multispectral remote sensing images. Hyperspectral image cube consist of three dimensions which the first and second dimensions related to the spatial domain and the third one related to the spectral domain which includes more than hundred bands. Most of the methods operate in the spectral domain for noise reduction while in this proposed method, a novel algorithm for reducing noise in hyperspectral images is implemented. The proposed method uses two different algorithms which are applied in two different hyperspectral images in both spatial and spectral domains. These images are Hyperion satellite image and AVIRIS airborne image. In order to reduce noise in the spatial domain, Total Variation (TV) algorithm and in the spectral domain, Wavelet algorithm is used. After the implementation of these methods, the results are fused at the pixel level. For the evaluation of the proposed method, the results were compared with other methods, both qualitatively and quantitatively. Various indices are used to assess the quantitative results which demonstrate the high accuracy of this method.CEI index for Hyperion image is 1.421 and for AVIRIS image is 0.0022. Another index is PSNR which the value for Hyperion image is 33.519 and for AVIRIS image is 22.371.