A New ADCS Method Based on Guided Filter for Tea HSIs

Abstract To improve the reconstructed performance of mass data storage, transmission and preserving spectral characteristics for tea hyperspectral images(HSIs), a new adaptive distributed compressive sensing method based on guided filter (ADCSGF) is proposed. According to the spectral characteristics of tea HSIs, all bands can be divided into different parts which can be further grouped with different band count. Bands of each group are compressed and reconstructed by distributed compressive sensing method, in which the adaptive bit stream allocation strategy based on the residual error is used to obtain the target bit rate for each non-key band and the key band is regarded as the guided filter band to improve the quality of reconstructed non-key bands in each group. The experimental results showed that ADCSGF can improve the subjective quality of image reconstruction and achieve at least a 1.5 dB higher peak signal-to-noise ratio (PSNR) of spectral dimension decorrelation method (SSDC) than that of distributed compressive sensing based on guided filter (DCSGF). ADCSGF can obtain better reconstructed spectral curve at the sampling rate of 0.1Bpp(Bytes per pixel) and achieve better normalized root mean square error (RMSE) performance at the sampling rate from 0.2Bpp to 0.5Bpp than those of DCSGF.