Decomposition-based multiscale compressed sensing

Multi-Scale Deep Compressive Sensing Network

“…As the proposed MS-DCSNet follows endto-end learning structure, inputs and labels are identical as the ground truth image. Similar to many image restoration methods [8][9][10][11][12][13][14], Euclidean loss is used as an objective function as:…”

“…Recent works [5][6][7][8] have proven that multi-scale CS can be the optimal sampling solution. Radial Fourier subsampling [1] can be considered as a simple example and often used in bioimaging due to its physically driven projection.…”

“…Another example is the block-based multi-scale sampling in MS-SPL [6] which first transforms image into wavelet domain, and then uses adaptive sample rate for each decomposition level. Canh et al developed multi-scale sampling for Kronecker CS in [7,8]. The authors first decomposed images into various scales by wavelet, pyramid, or multiple-resolution decomposition, and adaptively sampled each scale.…”

“…For example, CS cannot assume prior information on the to-be-sampled signal beyond general sparsity prior. Therefore, based on the very general observation of the human visual system, a low-frequency prior is typically assumed, which has motivated researchers to develop multiscale CS [5][6][7][8] which captures more low-frequency components in image/video signalsinput image signal is linearly decomposed into multi-scales and sampling is done adaptively to each scale.…”

“…However, most of the previous works were about a single scale sensing. While multiscale CS has shown significant performance gain over conventional CS though much attention [5][6][7][8], it still calls for further research on DCS for multi-scale sampling. All of these observations motivated this paper to exploit DL-based method for multi-scale sampling scheme.…”