Image compressed sensing recovery method based on iterative error compensation

Abstract

The invention discloses an image compressed sensing recovery method based on iterative error compensation, and belongs to the field of deep learning. According to the method, on the basis of a sampling and recovery unit of a traditional image compressed sensing method, an adder, a multiplier and a delayer are added, a closed-loop negative feedback structure is formed, image recovery error iterative compensation is carried out, and the image recovery quality is improved. Compared with ten latest image compression sensing algorithms based on deep learning on three standard data sets, the image restoration quality is obviously improved for different sampling rates from 0.01 to 0.50. The maximum increment of the peak signal to noise ratio (PSNR) can reach 4dB, and the maximum increment of the structural similarity (SSIM) can reach 0.03. The method can be applied to any existing image compressed sensing method, and the reconstruction capability is enhanced.

Description

Technical Field

[0001] This invention relates to the field of deep learning, and more particularly to an image compressed sensing restoration method. Background Technology

[0002] Image compressed sensing typically utilizes various prior knowledge to recover the original image. Classic methods include total variation, wavelet transform, sparse representation, low-rank representation, and deep learning.

[0003] Image compressed sensing technology can simultaneously sample and compress the source signal, employing a sub-Nyquist sampling frequency far below the Nyquist frequency. Leveraging known prior knowledge of the image, it can efficiently reconstruct the original image. This technology boasts a series of significant advantages, including low sampling rate, low power consumption, low system cost, minimal radiation damage, and high reconstruction performance. However, the extremely low sampling rate leads to the loss of a significant amount of useful information.

[0004] like Figure 1 As shown, traditional image compressed sensing systems have an open-loop structure, employing separate sampling units (MS) and reconstruction units (RC). Image compressed sensing strives to achieve an optimal balance between low sampling rates and high reconstruction capabilities, and its performance still has considerable room for improvement: the sampling rate is expected to be further reduced, and the image reconstruction quality is expected to be further improved. Summary of the Invention

[0005] Objective of the invention: To address the aforementioned existing technologies, this invention proposes an image compressed sensing restoration method based on iterative error compensation to improve image reconstruction quality.

[0006] Technical Solution: An image compressed sensing restoration method based on iterative error compensation, which restores the original image using a reconstructed sample y0. The method includes: inputting the reconstructed sample y0 and the output of the first delay unit DL into a comparator; inputting the error variable e output by the comparator into a multiplier and multiplying it with a constant multiplication factor λ to obtain a control variable c; inputting the control variable c and the output of the second delay unit DL in the (n-1)th iteration into the sample x. n-1 The input samples are fed into the adder, and the adder outputs the input sample x for the nth iteration. n x n Simultaneously input to the sampling unit MS and the second delay unit DL; the sampled sample z output by the sampling unit MS. n The input to the recovery unit RC is connected to the input of the first delay unit; the recovered image x is obtained when the system stabilizes. n .

[0007] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the computer program for the image compression-sensory restoration method.

[0008] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the image compressed sensing restoration method.

[0009] Beneficial Effects: This invention, based on the sampling and restoration units of traditional compressed sensing methods, incorporates adders, multipliers, and delayers to form a closed-loop negative feedback structure for iterative compensation of image restoration errors, thereby improving image restoration quality. Comparison with 10 state-of-the-art deep learning-based compressed sensing algorithms on three standard datasets (Set5, Set11, and Set14) shows significant improvements in image restoration quality for different sampling rates from 0.01 to 0.50. The maximum increase in Peak Signal-to-Noise Ratio (PSNR) is 4 dB, and the maximum increase in Structural Similarity Indicator (SSIM) is 0.03. This method can be applied to any existing compressed sensing method and enhances its reconstruction capabilities. Attached Figure Description

[0010] Figure 1 This is a block diagram of a traditional open-loop graph compressed sensing method; Figure 2 This is a block diagram of the closed-loop graph compressed sensing method of the present invention; Figure 3 The PSNR comparison results are for dataset Set5; Figure 4 The results of the SSIM comparison on dataset Set5. Detailed Implementation

[0011] The invention will now be further explained with reference to the accompanying drawings.

[0012] like Figure 2 As shown, an image compressed sensing restoration method based on iterative error compensation is proposed. Based on existing arbitrary image compressed sensing algorithms, the original image is restored by using reconstructed sample y0. The method consists of a closed-loop structure composed of a sampling unit MS, a restoration unit RC, an adder, a multiplier, and a delay unit DL. The sampling unit MS and the restoration unit RC constitute the controlled object.

[0013] Where e is the error variable, c is the control variable, λ is the constant multiplication factor, n is the iteration number, and x n Let x be the input sample for the nth iteration. n-1 Let y be the input sample for the (n-1)th iteration. n For the nth iteration output sample, y n-1 This is the output sample for the (n-1)th iteration.

[0014] The reconstructed sample y0 and the output of the first delay unit DL are input into a comparator. The error variable e output by the comparator is input into a multiplier and multiplied with a constant multiplication factor λ to obtain the control variable c. The control variable c and the output of the second delay unit DL in the (n-1)th iteration are input into the sample x. n-1 The input samples are fed into the adder, and the adder outputs the input sample x for the nth iteration. n x n Simultaneously input to the sampling unit MS and the second delay unit DL of the existing image compressed sensing method; the sampling sample z output by the sampling unit MS n The input to the recovery unit RC is connected to the input of the first delay unit. Its iterative mathematical formula is described as follows:

[0015] A closed-loop system recovers an ideal x0 from a known y0, with y0 as the input and x0 as the output. n The sampling unit MS and the recovery unit RC constitute the controlled object. The comparator is used to compare y... n-1 The difference between y0 and y0; the adder accumulates this difference. If y n First increase, then e decreases, c decreases, and x... n Decrease, eventually y n Decrease; and vice versa. After multiple comparisons and adjustments, the system tends to stabilize. According to negative feedback theory, after the system stabilizes, y n Approaching y0. Therefore, to ensure y n Approaching y0, x after the system stabilizes n It should also approximate x0. That is, when the closed-loop system is stable, an improved recovered image x is obtained. n .

[0016] The image reconstruction performance of our proposed method was compared with existing methods through experiments. The experimental software platform was the PyTorch deep learning framework on a 64-bit operating system; the experimental hardware platform was a laptop equipped with a 2.6 GHz dual-core processor and 8 GB of memory. The constant multiplication factor λ was set to 1, and the total number of iterations was set to 5. The 10 existing methods and related datasets used in the experiments are shown in Table 1. The number at the end of each dataset name indicates the total number of images contained in that dataset.

[0017] Table 1 Existing methods Dataset COAST Set5, Set11, Set14 MADUN Set5, Set11, Set14 ISTA Set5, Set11, Set14 ISTA+ Set5, Set11, Set14 ISTA++ Set5, Set11, Set14 OPINE Set5, Set11, Set14 AMP-NET Set5, Set11, Set14 MTC-CSNET Set5, Set11, Set14 TCS-NET Set5, Set11, Set14 TransCS Set5, Set11, Set14 The comparison structure of the PSNR and SSIM of the proposed method and existing methods on dataset Set5 is as follows: Figure 3 and 4 As shown, the proposed method outperforms traditional methods in image reconstruction.

[0018] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the above-described image compressed sensing restoration method based on iterative error compensation.

[0019] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the image compressed sensing restoration method based on iterative error compensation described above.

[0020] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0021] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0022] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. An image compressed sensing restoration method based on iterative error compensation, which restores the original image using reconstructed samples y0, characterized in that, The method includes: inputting the reconstructed sample y0 and the output of the first delay unit DL into a comparator; inputting the error variable e output by the comparator into a multiplier; multiplying e with a constant multiplication factor λ to obtain the control variable c; and inputting the control variable c and the output of the second delay unit DL in the (n-1)th iteration into the sample x. n-1 The input samples are fed into the adder, and the adder outputs the input sample x for the nth iteration. n x n Simultaneously input to the sampling unit MS and the second delay unit DL; the sampled sample z output by the sampling unit MS. n The input to the recovery unit RC is connected to the input of the first delay unit; the recovered image x is obtained when the system stabilizes. n .

2. The image compressed sensing restoration method based on iterative error compensation according to claim 1, characterized in that, The sampling unit MS and the recovery unit RC constitute a controlled object, and the mathematical formula for the iteration is: 。 3. The image compressed sensing restoration method based on iterative error compensation according to claim 1 or 2, characterized in that, The sampling unit MS and the restoration unit RC are sampling units and restoration units of any existing image compression sensing algorithm.

4. The image compressed sensing restoration method based on iterative error compensation according to claim 1 or 2, characterized in that, The value range of the constant multiplication factor λ is 0.8 to 1.

2.

5. The image compressed sensing restoration method based on iterative error compensation according to claim 1 or 2, characterized in that, The number of iterations is set to 5 to 10.

Images