Synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal

Abstract

The invention relates to a synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal. The synchronous denoising processing algorithm comprises the following steps: reading a remote sensing image to be denoised; constructing an independent image processing module by using an improved non-local mean algorithm and an EM algorithm to remove stripe noise and random noise in the image; a plurality of IMM modules are deployed in an assembly line form to realize multiple iterations of image data, and synchronous removal of image stripe noise and random noise is completed. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal is different from a stripe noise and random noise synchronous removal algorithm of a current software platform, and fully considers application requirements and processing logic of the hardware platform; and real-time synchronous denoising processing of remote sensing image data can be realized on hardware platforms such as an FPGA (Field Programmable Gate Array).

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image processing, and in particular to a synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal. Background Art

[0002] In order to meet the mission requirements of remote sensing detection, remote sensing cameras usually use swing-scan or push-scan mode for imaging. When using push-scan mode for imaging, the image data obtained by the camera usually contains obvious non-uniform stripes and random noise due to factors such as the detector manufacturing process, imaging mode, and ambient temperature, which seriously restricts the subsequent application of remote sensing data.

[0003] The existing synchronous denoising processing algorithms all have high computational complexity, and even on software platforms with huge computing power, they still require a long computing time, which is obviously difficult to meet the needs of rapid response of future remote sensing imaging. Summary of the invention

[0004] In order to effectively solve the problem that the existing image synchronous denoising algorithm has high computational complexity and is difficult to apply to the hardware platform to realize real-time preprocessing of onboard platform remote sensing data, the present invention proposes a synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal.

[0005] In order to solve the above technical problems, the technical solutions of the present invention are as follows:

[0006] A synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal includes the following steps:

[0007] Step 1: Read the remote sensing image to be denoised;

[0008] Step 2: Expand the bit width of the input pixel data;

[0009] The pixel data is stored in the high and low bits to form PIX-MAP. In the subsequent processing, only the low bit data is processed, and the high bit data is only used for data transmission.

[0010] Step 3: Cache the input PIX-MAP data through the first-level line cache, use the register to obtain the image window data, and pack it into WIN-MAP;

[0011] Step 4: Cache the data in WIN-MAP through the secondary row cache, use registers to obtain window data in the search domain, and pack the data to form SER-MAP;

[0012] Step 5: The image processing module decodes the data in the SER-MAP to obtain all window data in the search domain, and uses the improved non-local mean algorithm to calculate the estimated value of the target pixel to obtain the Estimated-MAP;

[0013] Step 6: Use the data in Estimated-MAP to estimate the stripe noise in the remote sensing image through the adjacent row and column mean algorithm;

[0014] Step 7: The real image information is updated by subtracting the original data stored in the high-order bit and the processed low-order bit in step 2, and the subtraction is used as the input for the next iteration, and the process returns to step 3 and iterates repeatedly until the loop end condition is met, and then step 8 is executed;

[0015] Step 8: Output the denoised image.

[0016] In the above technical solution, in step 1, the remote sensing image is a degraded image containing stripe noise and random noise.

[0017] In the above technical solution, in step 2, when processing the low-bit data, the 8-bit pixel data transmitted by the camera is expanded to 16-bit PIX-MAP data by copying itself once.

[0018] In the above technical solution, step three is specifically as follows: the data in PIX-MAP is cached by a line buffer (Line Buffer) composed of FIFO, and then the data therein is packaged into a window data packet (PIX_WIN) through a window register of size 3*3, and finally the PIX-WIN data constitutes a window data array WIN-MAP.

[0019] In the above technical solution, step four is specifically as follows: the data PIX-WIN in WIN-MAP is packaged by the line buffer (Line Buffer) and the window register which are also constituted by FIFO, and finally forms a search domain window data packet (SER-MAP).

[0020] In the above technical solution, step five is specifically as follows: a single data in SER-MAP is decoded to obtain nine windows in the search domain, and after calculating the weight of each window and the center pixel through Euclidean distance, the M-NLM algorithm is used to calculate the estimated value of the target pixel, and finally a pixel estimated value array (Estimated-MAP) is formed.

[0021] In the above technical solution, step six is ​​specifically as follows: the data in Estimated-MAP is used to estimate the stripe noise in the degraded image by continuously calculating the column mean of two adjacent rows of data to finally obtain the stripe noise residual data (RES-MAP).

[0022] In the above technical solution, in step seven, the loop termination condition is: the loop reaches a maximum number of times Iter≤5.

[0023] The present invention has the following beneficial effects:

[0024] The synchronous denoising processing algorithm based on stripe noise estimation and random noise removal on the hardware platform of the present invention is different from the synchronous removal algorithm of stripe noise and random noise on the current software platform. It fully considers the application requirements and processing logic of the hardware platform, and can realize real-time synchronous denoising processing of remote sensing image data on hardware platforms such as FPGA.

[0025] The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal of the present invention utilizes the maximum likelihood estimation theory (MLE) to construct a separate image processing module (IMM) through an improved non-local mean algorithm (M-NLM) and an EM algorithm to remove stripe noise and random noise in an image; it fully conforms to the processing logic of the hardware platform and can realize synchronous denoising processing of image data under low data delay; multiple IMM modules are deployed in the form of a pipeline to realize multiple iterations of image data, and finally complete the synchronous removal of image stripe noise and random noise.

[0026] The synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal of the present invention is deployed in such Figure 7 The Xilinx xc7k410tffg676 FPGA platform shown in the figure can effectively remove noise components in remote sensing images of different scenes and different spatial resolutions, which has high application value. Figure 8 (degraded image in actual imaging experiment) and Fig. 9 (degraded image under uniform light field), good processing effect can be achieved. The actual processing effect is as follows Fig.10 and Fig.11 In addition, when the system clock is 50Mhz, the frame rate can reach 162.77fps when processing 8-bit 512*512 image data. When facing remote sensing images of different scenes and different spatial resolutions, the noise components in the image can be effectively removed, which has strong robustness.

[0027] The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal of the present invention is of great significance for improving the quality of remote sensing images and enhancing the image preprocessing capability of satellite platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0029] Figure 1 It is a schematic diagram of the overall process of the synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal of the present invention;

[0030] Figure 2 Schematic diagram of data bit width expansion

[0031] Figure 3 This is a schematic diagram of the first-level row cache;

[0032] Figure 4 This is a schematic diagram of the secondary row cache;

[0033] Figure 5 Schematic diagram of the operation of the improved non-local means algorithm (M-NLM);

[0034] Figure 6 Schematic diagram of the adjacent row and column mean (ACM) calculation method;

[0035] Figure 7 A schematic diagram of the hardware platform for deploying the framework;

[0036] Figure 8 It is a remote sensing image affected by both stripe noise and random noise in actual imaging experiments;

[0037] Fig. 9 It is a remote sensing image affected by stripe noise and random noise under uniform light field;

[0038] Fig.10 for Figure 8 The remote sensing image shown is a remote sensing image processed by the present invention;

[0039] Fig.11 for Fig. 9 The remote sensing image shown is a remote sensing image processed by the present invention. DETAILED DESCRIPTION

[0040] The inventive concept of the present invention is:

[0041] The invention discloses a synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal, which utilizes maximum likelihood estimation theory (MLE) to construct a separate image processing module (IMM) through an improved non-local mean algorithm (M-NLM) and an EM algorithm to remove stripe noise and random noise in an image, thereby realizing synchronous removal of stripe noise and random noise in remote sensing images.

[0042] The Chinese meanings of the English abbreviations in the following sections are:

[0043] PIX-MAP: pixel data array after bit width expansion;

[0044] WIN-MAP: Window data packet formed after the first-level line cache;

[0045] SER-MAP: Search domain window data packet formed after the secondary row cache;

[0046] IMM: Image processing module;

[0047] M-NLM: Improved non-local means algorithm;

[0048] Estimated-MAP: pixel estimation value array;

[0049] ACM algorithm: adjacent row and column mean algorithm;

[0050] RES-MAP: noisy residual data;

[0051] FPGA: Field Programmable Gate Array;

[0052] MLE: Maximum Likelihood Estimation Theory;

[0053] EM algorithm: expectation maximization algorithm.

[0054] The technical solution of the present invention will be described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0055] like Figure 1 As shown, the synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal of the present invention comprises the following specific steps:

[0056] Step 1: Read the remote sensing image to be denoised;

[0057] Step 2: Data bit width expansion to form PIX-MAP;

[0058] Step 3: cache the data in the first level row, and use registers to pack the window data to form a WIN-MAP;

[0059] Step 4: Perform secondary row cache on WIN-MAP data and use registers to pack window data in the search domain to form SER-MAP;

[0060] Step 5: The image processing module (IMM) decodes the data in the SER-MAP and uses the improved non-local mean algorithm (M-NLM) to calculate the estimated value of the target pixel to obtain the Estimated-MAP;

[0061] Step 6: Use the data in Estimated-MAP to estimate the stripe noise in the degraded image through the adjacent row and column mean algorithm (ACM algorithm);

[0062] Step 7: Update the input in step 3 and iterate;

[0063] Step 8: Output denoising results.

[0064] The synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal in this embodiment has the following specific process:

[0065] Step 1: Read the remote sensing image to be denoised, which is a degraded image containing stripe noise and random noise.

[0066] For the acquired remote sensing images, the following degradation model is satisfied:

[0067] D=G+S+N(0.1)

[0068] Where D∈R m×n , G∈R m×n , S∈R m×n and N∈R m×n Represent the degraded image, real image, stripe noise and Gaussian noise respectively.

[0069] Step 2: Expand the bit width of the input pixel data, register the pixel data in the high and low bits respectively to form a PIX-MAP, and only process the low-bit data in the subsequent processing, and only transfer the high-bit data.

[0070] In this step, the 8-bit pixel data downloaded by the camera is copied once by itself, so that the 9th bit is the same as the 1st bit (Bit0), the 10th bit is the same as the 2nd bit (Bit1), ..., the 16th bit is the same as the 8th bit (Bit7), such as Figure 2 As shown, the bit width is expanded to 16-bit PIX-MAP data.

[0071] Step 3: Cache the input PIX-MAP data through the first-level line buffer, use the register to obtain the image window data, and pack the data into WIN-MAP; in this step, the data in PIX-MAP is cached by the line buffer (Line Buffer) composed of FIFO, and then the data is packed into a window data packet (PIX_WIN) through a window register of size 3*3, and finally the PIX-WIN data constitutes the window data array WIN-MAP.

[0072] Step 4: Cache the data in WIN-MAP through the secondary row cache, use registers to obtain window data in the search domain, and pack the data to form SER-MAP.

[0073] In this step, the data PIX-WIN in WIN-MAP will be packaged by the line buffer (LineBuffer) and the window register which are also constituted by FIFO, and finally form a search domain window data packet (SER-MAP).

[0074] Step 5: The image processing module (IMM) decodes the data in the SER-MAP to obtain all window data in the search domain, and uses the improved non-local mean algorithm (M-NLM) to calculate the estimated value of the target pixel to obtain the Estimated-MAP;

[0075] In this step, a single data in SER-MAP is decoded to obtain nine windows in the search domain. After calculating the weight of each window and the center pixel through Euclidean distance, the Figure 5 The M-NLM algorithm shown calculates the estimated value of the target pixel, and finally forms an estimated pixel value array (Estimated-MAP).

[0076] Step 6: Use the adjacent row and column mean (ACM) algorithm to estimate the stripe noise in the degraded image using the data in the Estimated-MAP.

[0077] In this step, the data in the Estimated-MAP is used to estimate the stripe noise in the degraded image by continuously calculating the column mean of two adjacent rows of data, and finally obtain the stripe noise residual data (RES-MAP).

[0078] Step 7: The real image information is updated by subtracting the original data stored in the high-order register in step 2 from the data in the low-order processed RES-MAP, and using it as the input for the next iteration. Return to step 3 and iterate in a loop until the loop end condition is met, and then execute step 8.

[0079] In this step, the real image information is updated by subtracting the original data stored in the high-order register in step 2 from the estimated stripe noise in RES-MAP and using it as the input for the next iteration. Steps 3 to 7 are repeated until the termination condition Iter≤5 is met and the synchronous denoised image is output.

[0080] Step 8: Output the denoised image.

[0081] The synchronous denoising processing algorithm based on stripe noise estimation and random noise removal on the hardware platform of the present invention is different from the synchronous removal algorithm of stripe noise and random noise on the current software platform. It fully considers the application requirements and processing logic of the hardware platform, and can realize real-time synchronous denoising processing of remote sensing image data on hardware platforms such as FPGA.

[0082] The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal of the present invention utilizes the maximum likelihood estimation theory (MLE) to construct a separate image processing module (IMM) through an improved non-local mean algorithm (M-NLM) and an EM algorithm to remove stripe noise and random noise in an image; it fully conforms to the processing logic of the hardware platform and can realize synchronous denoising processing of image data under low data delay; multiple IMM modules are deployed in the form of a pipeline to realize multiple iterations of image data, and finally complete the synchronous removal of image stripe noise and random noise.

[0083] The synchronous denoising processing algorithm based on hardware platform strip noise estimation and random noise removal of the present invention is deployed in such Figure 7 The Xilinx xc7k410tffg676 FPGA platform shown in the figure can effectively remove noise components in remote sensing images of different scenes and different spatial resolutions, which has high application value. Figure 8 (degraded image in actual imaging experiment) and Fig. 9 (degraded image under uniform light field), good processing effect can be achieved. The actual processing effect is as follows Fig.10 and Fig.11 In addition, when the system clock is 50Mhz, the frame rate can reach 162.77fps when processing 8-bit 512*512 image data. When facing remote sensing images of different scenes and different spatial resolutions, the noise components in the image can be effectively removed, which has strong robustness.

[0084] The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal of the present invention is of great significance for improving the quality of remote sensing images and enhancing the image preprocessing capability of satellite platforms.

[0085] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the invention.

Claims

1. A synchronous denoising algorithm based on hardware platform strip noise estimation and random noise removal, characterized in that: The following steps are involved: Step 1: Read the remote sensing image to be denoised; Step 2: Expand the bit width of the input pixel data; The pixel data is stored in the high and low bits to form PIX-MAP. In the subsequent processing, only the low bit data is processed, and the high bit data is only used for data transmission. Step 3: Cache the input PIX-MAP data through the first-level line cache, use the register to obtain the image window data, and pack it into WIN-MAP; Step 4: Cache the data in WIN-MAP through the secondary row cache, use registers to obtain window data in the search domain, and pack the data to form SER-MAP; Step 5: The image processing module decodes the data in the SER-MAP to obtain all window data in the search domain, and uses the improved non-local mean algorithm to calculate the estimated value of the target pixel to obtain the Estimated-MAP; Step 6: Use the data in Estimated-MAP to estimate the stripe noise in the remote sensing image through the adjacent row and column mean algorithm; Step 7: The real image information is updated by subtracting the original data stored in the high-order bit and the processed low-order bit in step 2, and the subtraction is used as the input for the next iteration, and the process returns to step 3 and iterates repeatedly until the loop end condition is met, and then step 8 is executed; Step 8: Output the denoised image.

2. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: In step 1, the remote sensing image is a degraded image containing stripe noise and random noise.

3. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: In step 2, when processing the low-bit data, the 8-bit pixel data transmitted by the camera is expanded to 16-bit PIX-MAP data by copying itself once.

4. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: Step three is as follows: the data in PIX-MAP is cached by a row buffer formed by FIFO, and then the data is packaged into a window data packet through a window register of size 3*3, and finally the PIX-WIN data constitutes a window data array WIN-MAP.

5. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: Step 4 is as follows: the data PIX-WIN in WIN-MAP is packaged by the line buffer and window register which are also constituted by FIFO, and finally forms a search domain window data packet.

6. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: Step five is as follows: a single data in SER-MAP is decoded to obtain nine windows in the search domain. After calculating the weight of each window and the center pixel through Euclidean distance, the M-NLM algorithm is used to calculate the estimated value of the target pixel, and finally a pixel estimation value array is formed.

7. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: Step six is ​​as follows: the data in the Estimated-MAP is used to estimate the stripe noise in the degraded image by continuously calculating the column mean of two adjacent rows of data, and finally obtaining the stripe noise residual data.

8. The synchronous denoising processing algorithm based on hardware platform stripe noise estimation and random noise removal according to claim 1 is characterized in that: In step seven, the loop termination condition is: the loop reaches a maximum number of times Iter≤5.

Images