Wavelet denoising algorithm based on OCT image

Abstract

The invention relates to the technical field of image denoising, in particular to a wavelet denoising algorithm based on an OCT image, which comprises three stages of wavelet decomposition, threshold quantization and reconstruction, the wavelet decomposition comprises low-frequency sub-signals, high-frequency sub-signals in the horizontal direction, high-frequency sub-signals in the vertical direction and high-frequency sub-signals in the diagonal direction, the high-frequency sub-signals in the horizontal direction, the high-frequency sub-signals in the vertical direction and the high-frequency sub-signals in the diagonal direction are processed by adopting different mean filtering templates, and the high-frequency sub-signals in the horizontal direction are filtered by using a horizontal template so as to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction. According to the wavelet denoising algorithm based on the OCT image, the image is subjected to wavelet decomposition, high-frequency components are denoised through mean filtering, then wavelet fusion is carried out, a good effect is achieved, and edge details of the image are protected while noisy points are removed to the maximum extent.

Description

Technical Field

[0001] The present invention relates to the technical field of image denoising, and in particular to a wavelet denoising algorithm based on OCT images. Background Art

[0002] OCT imaging technology is an imaging technology commonly used in the medical field and plays an important role in treatment and diagnosis.

[0003] In the acquisition and transmission of OCT endoscope images, they are usually affected by various noises, which has an adverse effect on subsequent image recognition, doctor diagnosis and other processes. Using traditional denoising methods: simple denoising algorithms such as median filtering and mean filtering can effectively filter out the influence of noise, but this will also cause the loss of image edge information. Wavelet transform shows excellent local properties and can efficiently capture edge details in images. However, most of the current wavelet denoising technologies only use one wavelet basis, which makes it difficult to achieve an ideal balance between the smoothing of the edge and texture parts of the image. Summary of the invention

[0004] The object of the present invention is to provide a wavelet denoising algorithm based on OCT images to solve the problem that most of the current wavelet denoising technologies proposed in the above background technology only use one wavelet basis, which makes it difficult to achieve an ideal balance between the smoothing of the edge and texture parts of the image. To achieve the above object, the present invention provides the following technical solution: a wavelet denoising algorithm based on OCT images, including three stages of wavelet decomposition, threshold quantization and reconstruction, the wavelet decomposition includes a low-frequency sub-signal, a high-frequency sub-signal in the horizontal direction, a high-frequency sub-signal in the vertical direction and a high-frequency sub-signal in the diagonal direction, the high-frequency sub-signal in the horizontal direction, the high-frequency sub-signal in the vertical direction and the high-frequency sub-signal in the diagonal direction are processed using different mean filter templates, the horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction, the horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction, and the diagonal high-frequency sub-signal is processed using a diagonal template to eliminate high-frequency noise in the horizontal and vertical directions.

[0005] Further preferably, the threshold estimation method adopts the VisuShrink threshold, and the VisuShrink threshold formula is:

[0006]

[0007] Where T is the threshold, σ represents the standard deviation of the noise, and M*N represents the length and width of the image. The standard deviation of the noise is obtained by the following formula:

[0008]

[0009] Where Det represents the wavelet coefficient of the minimum layer, and median(·) represents the median;

[0010] Set the initial wavelet coefficient to ω jk , the wavelet coefficient after threshold processing is T is the threshold, then the hard threshold function:

[0011]

[0012] Soft threshold function:

[0013]

[0014] Where sgn(·) is the sign function.

[0015] The mathematical meaning of the mean filter is as follows:

[0016]

[0017] Where S is the area set centered at (x, y), and M is the number of pixels in S.

[0018] The mean filter template is as follows:

[0019]

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] In the present invention, the image is decomposed by wavelet, and the high frequency component is denoised by mean filtering, and then wavelet fusion is performed, so as to achieve a better effect, and protect the edge details of the image while removing the noise to the maximum extent.

[0022] In the present invention, based on this algorithm, a three-dimensional reconstruction algorithm can be studied for the denoised OCT image. The three-dimensional reconstruction of the image can help doctors accurately locate the lesion site of the patient and provide a basis for clinical diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flowchart of the algorithm of the present invention;

[0024] Figure 2 It is a hardware flow chart of the present invention;

[0025] Figure 3 This is a comparison diagram of image effects in the first embodiment of the present invention;

[0026] Figure 4 is a sample image data diagram of the present invention;

[0027] Figure 5This is a comparison diagram of the image effects of the second embodiment of the present invention;

[0028] Figure 6 This is a comparison chart of EPI data of the present invention. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technical personnel in this field without creative work are within the scope of protection of the present invention.

[0030] See also Figure 1-Figure 2 The present invention provides a technical solution: including three stages: wavelet decomposition, threshold quantization and reconstruction. The wavelet decomposition includes a low-frequency sub-signal, a high-frequency sub-signal in the horizontal direction, a high-frequency sub-signal in the vertical direction and a high-frequency sub-signal in the diagonal direction. The high-frequency sub-signal in the horizontal direction, the high-frequency sub-signal in the vertical direction and the high-frequency sub-signal in the diagonal direction are processed using different mean filter templates. The horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction. The horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction. The diagonal high-frequency sub-signal is processed using a diagonal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction.

[0031] Further preferably, the threshold estimation method adopts the VisuShrink threshold, and the VisuShrink threshold formula is:

[0032]

[0033] Where T is the threshold, σ represents the standard deviation of the noise, and M*N represents the length and width of the image. The standard deviation of the noise is obtained by the following formula:

[0034]

[0035] Where Det represents the wavelet coefficient of the minimum layer, and median(·) represents the median;

[0036] Set the initial wavelet coefficient to ω jk , the wavelet coefficient after threshold processing is T is the threshold, then the hard threshold function:

[0037]

[0038] Soft threshold function:

[0039]

[0040] Where sgn(·) is the sign function.

[0041] The mathematical meaning of the mean filter is as follows:

[0042]

[0043] Where S is the area set centered at (x, y), and M is the number of pixels in S.

[0044] The mean filter template is as follows:

[0045]

[0046] The algorithm is deployed on the FPGA development board, and the haar function is used for forward wavelet transform. First, the column transform is performed, that is, the haar function performs sliding convolution along every two pixels in each row, which can divide the image into two parts, high-pass and low-pass, called H / L. Then the row transform is performed to further divide the two parts into four parts, HH / HL / LH / LL. Among these four parts, HH / HL / LH is mean filtered to remove noise. LL does not need to be filtered but needs to be synchronized with the first three parts in terms of timing. After mean filtering, HH / HL / LH / LL is inversely wavelet transformed to restore the original image. Since the SD card reading rate and HDMI display rate do not match, the pixel data read by the SD needs to be temporarily stored in the BRAM, and the HDMI module is triggered to read from the BRAM after the complete reading.

[0047] The edge protection index (EPI) and image entropy are used to evaluate the image. Since OCT images are affected by speckle noise during the acquisition process, image entropy is used as a test indicator. The closer the edge protection index is to 1, the better the algorithm protects the edge of the image. The larger the image entropy is, the clearer the image is. The calculation formula is as follows:

[0048]

[0049] where s and Represent the original image data and the denoised image data respectively, Δ is the Laplace operator, and:

[0050] Γ(S1, S2)= (x,y) S1(x,y)S2(x,y);

[0051] And Pi is the grayscale frequency of the image in the area, i is the grayscale frequency of the image

[0052] degree value, and stipulate that log2(0)=0.

[0053] Embodiment 1:

[0054] The OCT endoscope images with a pixel size of 512X512 were used, speckle noise with variances of 0.02 and 0.04 were added, and a 1X3 template was used for the horizontal component, a 3X1 template was used for the vertical component, and a 5X5 template was used for the diagonal component. Figure 3 As shown:

[0055] (a) shows the original image, (b) and (c) show the original image with 0.02 speckle noise added and the denoised image, respectively, (d) and (e) show the original image with 0.04 speckle noise added and the denoised image, respectively.

[0056] Embodiment 2:

[0057] The OCT endoscope image with a pixel size of 512X512 is used for noise reduction using mean filtering, median filtering and the algorithm proposed in the application, such as Figure 5 As shown:

[0058] (f) shows the original image, (g) is a noisy image with speckle noise added, (h) shows the image after mean filtering, (i) shows the image after median filtering, and (j) shows the image after denoising using the algorithm in this paper.

[0059] By observing the image results, it can be seen that the traditional algorithm has a good overall visual effect, but performs poorly in protecting edge details, the noise removal effect is not significant enough, and the overall effect is poor; in contrast, the algorithm proposed in the present application performs better in denoising, can effectively suppress noise and significantly improve edge protection capabilities, and the denoised image has good smoothness. It can be seen from the table that compared with the traditional denoising method, the denoised image of the application has an EPI value closer to 1 and a higher image entropy, which shows that while protecting edge information, the image is displayed more clearly, objectively illustrating the superiority of the denoising effect. In addition, the image filter template size used in the algorithm of the present application can be freely adjusted according to the noise intensity in the image, and multi-scale wavelet decomposition is used when the noise is strong to obtain better denoising effect.

[0060] The above shows and describes the basic principles, main features and advantages of the present invention. Technical personnel in this industry should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. Wavelet denoising algorithm based on OCT images, characterized in that: The method comprises three stages: wavelet decomposition, threshold quantization and reconstruction. The wavelet decomposition comprises a low-frequency sub-signal, a high-frequency sub-signal in the horizontal direction, a high-frequency sub-signal in the vertical direction and a high-frequency sub-signal in the diagonal direction. The high-frequency sub-signal in the horizontal direction, the high-frequency sub-signal in the vertical direction and the high-frequency sub-signal in the diagonal direction are processed using different mean filter templates. The horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction. The horizontal high-frequency sub-signal is filtered using a horizontal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction. The diagonal high-frequency sub-signal is processed using a diagonal template to eliminate high-frequency noise in the horizontal direction and retain low-frequency information in the vertical direction.

2. The wavelet denoising algorithm based on OCT images according to claim 2 is characterized in that: The threshold estimation method adopts the VisuShrink threshold, and the VisuShrink threshold formula is: Where T is the threshold, σ represents the standard deviation of the noise, and M*N represents the length and width of the image. The standard deviation of the noise is obtained by the following formula: Where Det represents the wavelet coefficient of the minimum layer, and median(·) represents the median; Set the initial wavelet coefficient to ω jk , the wavelet coefficient after threshold processing is T is the threshold, then the hard threshold function: Soft threshold function: Where sgn(·) is the sign function.

3. The wavelet denoising algorithm based on OCT images according to claim 1, characterized in that: The mathematical meaning of the mean filter is as follows: Where S is the area set centered at (x, y), and M is the number of pixels in S.

4. The wavelet denoising algorithm based on OCT image according to claim 1 is characterized in that: The mean filter template is as follows:

Images