TWI459795B - Noise reduction method - Google Patents

Description

Noise suppression method

The present invention relates to an image processing method, and more particularly to a method for eliminating image noise.

The advancement of multimedia technology has made modern people's demands for high-quality images increasingly. The quality of the image is quite related to the amount of noise accompanying the capture of images, signal conversion and transmission. In order to effectively remove noise to improve image quality, research on noise cancellation in the field of image processing has received more and more attention. Common noise cancellation methods can be divided into space-based de-noising processing methods and time-based de-noising processing methods.

The space-based de-noising processing method is mainly for processing a single image. However, since the details such as edges or textures in the original image are often destroyed during processing, it is easy to produce a blurring result after removing noise.

The time-based de-noising processing method utilizes multiple images for processing. Although the details in the image can be preserved, it is easy to generate a moving afterimage when filtering the moving objects in the image, so that it is easy to Discomfort caused by viewing. In addition, the processing with multiple images is highly complex and computationally intensive, so it is difficult to meet the need for instant noise removal.

In view of the above, the present invention provides a noise suppression method for more effectively eliminating noise in a source image through two-stage denoising processing.

The invention provides a noise suppression method for filtering noise in a source image by using a reference image. The source image and the reference image respectively comprise n 2×2 pixel blocks, and each pixel in each 2×2 pixel block has a corresponding red channel according to a Bayer pattern color arrangement rule (R color channel) a pixel value of one of a first green channel (Gr color channel), a second green channel (Gb color channel), and a blue color channel (B color channel), and n is a positive integer. The method includes generating a first quasi-image of the corresponding source image and a second quasi-image of the corresponding reference image. The first pseudo-image includes n source pseudo-pixels, and each source pseudo-pixel corresponds to a 2×2 pixel block in the source image. The second pseudo-image includes n reference pseudo-pixels, and each reference quasi-pixel corresponds to a 2×2 pixel block in the reference image. The method further includes calculating a global motion vector between the first pseudo image and the second pseudo image, performing dynamic compensation processing on the reference image according to the global motion vector to obtain a plurality of dynamic compensation results, and performing the source image by using the dynamic compensation result. Temporal blending processing to generate a first-order noise suppression image, and spatial noise reduction processing on the first-order noise suppression image to generate a second-order noise suppression image.

In an embodiment of the present invention, the step of performing dynamic compensation processing on the reference image according to the global motion vector to obtain the plurality of dynamic compensation results includes aligning the reference image according to the global motion vector to generate the corresponding red channel, respectively. a first dynamic compensation result, a second dynamic compensation result, a third dynamic compensation result, and a fourth dynamic compensation result of a green channel, a second green channel, and a blue channel.

In an embodiment of the invention, the step of performing time superimposition processing on the source image by using the dynamic compensation result to generate the first-order noise suppression image further includes quasi-pixels for each source, in all reference pseudo-pixels. Find a reference quasi-pixel that has a corresponding position to the source pseudo-pixel. According to the color and brightness difference between the source pseudo-pixel and the reference quasi-pixel, it is judged whether the 2×2 pixel block corresponding to the source pseudo-pixel is time-superimposed. To perform time superimposition processing, each pixel value in the 2×2 pixel block corresponding to the source pseudo pixel is matched with the first dynamic compensation result, the second dynamic compensation result, the third dynamic compensation result, and the fourth The pixel values at the corresponding positions in the dynamic compensation result are superimposed to produce a partial overlap result. If the time superimposition process is not performed, the 2×2 pixel block corresponding to the source pseudo-pixel is used as the partial superposition result. Finally, according to the Belle color arrangement rule, the partial superposition results corresponding to the pseudo-pixels of each source are reconstructed, thereby generating the first-order noise suppression image.

In an embodiment of the invention, the step of determining whether to perform time superimposition processing on the 2×2 pixel block corresponding to the source pseudo pixel according to the color and brightness difference between the source pseudo pixel and the reference quasi pixel includes: The absolute difference between the pixel value of the source pseudo-pixel and the pixel value of the corresponding blue channel in the 2×2 pixel block corresponding to the source pseudo-pixel is defined as a source pseudo-blue chromaticity. The absolute difference between the pixel value of the source pseudo-pixel and the pixel value of the corresponding red channel in the 2×2 pixel block corresponding to the source pseudo-pixel is defined as a source pseudo-red chromaticity. The absolute difference between the pixel value of the reference pseudo pixel and the pixel value of the corresponding blue channel in the 2×2 pixel block corresponding to the reference pseudo pixel is defined as a reference pseudo blue chromaticity. The absolute difference between the pixel value of the reference pseudo-pixel and the pixel value of the corresponding red channel in the 2×2 pixel block corresponding to the reference pseudo-pixel is defined as a reference pseudo-red chromaticity. And, a luminance threshold value, a blue chrominance threshold value, and a red chrominance threshold value corresponding to the source pseudo-pixel are obtained. When the absolute difference between the pixel value of the source pseudo-pixel and the pixel value of the reference pseudo-pixel is less than the luminance threshold, the absolute difference between the source pseudo-blue chromaticity and the reference pseudo-blue chromaticity is less than the blue chrominance threshold, and the source is pseudo-red When the absolute difference between the degree and the reference pseudo-red chromaticity is less than the red chrominance threshold, it is determined that the 2×2 pixel block corresponding to the source pseudo-pixel is time-superimposed.

In an embodiment of the invention, the step of obtaining a luminance threshold value corresponding to the source pseudo-pixel includes obtaining, in the first pseudo-image, an m×m pixel block centered on the source pseudo-pixel, wherein m is a positive integer. Calculate the Mean Absolute Error (MAE) of each pixel value in the m×m pixel block. An initial luminance value is obtained according to the average absolute error, and a luminance gain value is obtained according to the pixel value of the source pseudo pixel, and the product of the luminance initial value and the luminance gain value is used as the luminance threshold of the corresponding source pseudo pixel.

In an embodiment of the invention, the step of obtaining a blue chrominance threshold corresponding to the source pseudo-pixel comprises obtaining a blue chrominance gain value based on the source pseudo-blue chromaticity. Then, the product of the luminance initial value and the blue chrominance gain value is used as the blue chrominance threshold of the corresponding source pseudo-pixel.

In an embodiment of the invention, the step of obtaining a red chrominance threshold corresponding to the source pseudo-pixel includes obtaining a red chrominance gain value according to the source pseudo-red chromaticity, and using the luminance initial value and the red chrominance gain value The product is used as the red chrominance threshold for the pixel of the corresponding source.

In an embodiment of the invention, the step of generating a partial overlay result corresponding to the source pseudo-pixel further comprises defining a first weight preset value and a second weight preset value. In the first pseudo-image, an m×m pixel block is obtained centering on the source pseudo-pixel, where m is a positive integer, and the average absolute error of each pixel value in the m×m pixel block is calculated. A time overlap weight value is obtained according to the average absolute error, and a partial overlap result corresponding to the source pseudo pixel is generated according to the time overlap weight value.

In an embodiment of the invention, the step of generating a partial superposition result corresponding to the source pseudo-pixel according to the time-stacked weight value comprises calculating by the formula (1):

Where BR represents the local superposition result of the corresponding source pseudo-pixel. Wc represents the time overlap weight value, and C_base represents the preset coefficient. Rs, Grs, Gbs, and Bs respectively represent pixel values of the corresponding red channel, the first green channel, the second green channel, and the blue channel in the 2×2 pixel block corresponding to the source pseudo pixel. And Ra, Gra, Gba, and Ba respectively indicate that in the first dynamic compensation result, the second dynamic compensation result, the third dynamic compensation result, and the fourth dynamic compensation result, in the 2×2 pixel block corresponding to the source pseudo pixel Each pixel is at the pixel value of the corresponding position.

Based on the above, the present invention uses at least one reference image to obtain dynamic compensation results corresponding to different color channels when performing noise cancellation on the source image conforming to the Belle color arrangement rule, and uses the dynamic compensation result to perform the source image first. The time overlap processing is performed, and then the result of the time superimposition processing is spatially denoised, thereby effectively eliminating the negative influence of the noise on the source image.

The above described features and advantages of the present invention will be more apparent from the following description.

FIG. 1 is a flow chart of a method for suppressing noise according to an embodiment of the invention. The noise suppression method in this embodiment uses at least one reference image to filter out noise in the source image. The method is applicable to a charge coupled device (CCD) or a complementary metal oxide semiconductor (Complementary). A digital image capturing device such as a digital camera or a digital camera, such as a metal-Oxide semiconductor (CMOS). Since the image sensing component can only detect the intensity of the light and cannot distinguish the wavelength of the light, the color cannot be distinguished. Therefore, a Bayer filter needs to be installed in front of the image sensing component to filter out the red light, Blue light and green light can achieve the intensity of red, blue and green. In other words, in the source image from which the noise is to be removed in this embodiment, each pixel has a corresponding red channel (R color channel) and a first green channel (Gr color channel according to the Bayer pattern color arrangement rule). ), the pixel value of one of the second green channel (Gb color channel) and the blue color channel (B color channel).

Taking the source image 200 shown in FIG. 2 as an example, the source image 200 has 8×6 pixels, and the pixels can be divided into 12 pixel blocks (eg, pixel blocks 210_a, 220_a) having a size of 2×2. Each pixel block includes four pixels, wherein the upper left pixel has a pixel value corresponding to the first green channel (labeled as Gr in FIG. 2), and the upper right pixel has a pixel value corresponding to the red channel (labeled as R in FIG. 2) The lower left pixel has a pixel value corresponding to the blue channel (labeled B in FIG. 2), and the lower right pixel has a pixel value corresponding to the second green channel (labeled as Gb in FIG. 2). In other embodiments, the pixel values of the upper left, upper right, lower left, and lower right pixels included in each pixel block may also correspond to the red channel, the first green channel, the second green channel, and the blue channel, respectively.

In this embodiment, the reference image and the source image for assisting in filtering noise have the same size. Assuming that the source image includes n 2×2 pixel blocks (n is a positive integer), the reference image also includes n 2×2 pixel blocks, and each pixel in each pixel block of the reference image is also in accordance with the Bell diagram. The color arrangement rules have pixel values corresponding to one of the red channel, the first green channel, the second green channel, and the blue channel.

The detailed steps of filtering the noise of the source image using the reference image will be described below. First, as shown in step S210, a first pseudo image corresponding to the source image and a second pseudo image corresponding to the reference image are generated. If the source image includes n 2×2 pixel blocks, the first quasi-image includes n source pseudo-pixels, and each source pseudo-pixel corresponds to a 2×2 pixel block in the source image. The second pseudo-image includes n reference pseudo-pixels, and each reference quasi-pixel corresponds to a 2×2 pixel block in the reference image. In other words, if the size of the source image and the reference image is W×H, the size of the first pseudo image and the second pseudo image may be (W/2)×(H/2).

Taking the source image 200 shown in FIG. 2 as an example, the first pseudo image 300 generated corresponding to the source image 200 is as shown in FIG. 3. Referring to FIG. 2 and FIG. 3 , the source pseudo-pixel 210_b in the first pseudo-image 300 corresponds to the pixel block 210_a in the source image 200, and the source pseudo-pixel 220_b in the first quasi-image 300 is in the corresponding source image 200. Pixel block 220_a, and so on. In this implementation, the pixel value of each source of the first pseudo-image 300 may be an average of four pixel values in the corresponding pixel block. For example, the pixel value of the source pseudo-pixel 210_b is the pixel value average of the four pixels of the upper left, upper right, lower left, and lower right included in the pixel block 210_a.

Since each of the source image and the reference image corresponds to one of blue, red, and green, respectively, the source image and the reference image cannot be directly used to calculate the motion vector required for the motion compensation process to be performed later. In this regard, the present embodiment will utilize the first and second pseudo-images to calculate the motion vector. As shown in step S120, a global motion vector between the first pseudo image and the second quasi image is calculated. Next, in step S130, the reference image is subjected to dynamic compensation processing according to the global motion vector, thereby obtaining a plurality of dynamic compensation results.

The present invention does not limit the manner in which global motion vectors are computed. In the case of dynamic compensation processing, the reference image is first ordered to align the source image according to the global motion vector. Specifically, assuming that the global motion vector is (p, q), the reference image is shifted to the right by p pixels and moved downward by q pixels. Then, the pixel values of all corresponding red channels are separated from the alignment result to generate a first dynamic compensation result corresponding to the red channel. Similarly, the pixel values corresponding to the first green channel, the second green channel, and the blue channel are respectively separated from the alignment result, and the second dynamic compensation result corresponding to the first green channel is generated, corresponding to the second green channel. The third dynamic compensation result and the fourth dynamic compensation result corresponding to the blue channel.

Next, in step S140, temporal fusion processing is performed on the source image by using the dynamic compensation result to generate a first-order noise suppression image. In this embodiment, the feature details and brightness of the source image are taken into consideration when performing the time superimposition process, and then the different weight values are used for superposition. The detailed steps for generating the first-order noise suppression image will be described later in conjunction with the illustration.

Finally, as shown in step S150, spatial noise reduction processing is further performed on the first-order noise suppression image to generate a second-order noise suppression image. The present invention does not limit the manner in which spatial noise removal processing is performed.

As shown in FIG. 1 , in this embodiment, the effect of three-dimensional noise removal is achieved by performing time superimposition processing and spatial de-noise processing. Therefore, for a digital image capturing device having a high frame rate and a high resolution, the steps in the image captured by the method can be more effectively eliminated by the steps shown in FIG. News, and thus improve image quality.

The detailed steps of generating a first-order noise suppression image by using a plurality of dynamic compensation results to perform time superimposition processing on the source image in step S140 of FIG. 1 will be described below with reference to FIG. In short, in this embodiment, a partial superposition result of 2×2 pixels is generated for each source pseudo pixel, and then the partial superposition result corresponding to each source pseudo pixel is used to generate the first. Order noise suppresses the image.

Referring to FIG. 4, first, as shown in step S410, a source pseudo-pixel is obtained from all source pseudo-pixels included in the first pseudo-image. For example, in this embodiment, the source pseudo-pixels are sequentially obtained according to the pixel position. For convenience of explanation, the following (x, y) indicates the pixel position of the source pseudo-pixel in the first pseudo-image to be processed.

Next, as shown in step S420, a reference quasi-pixel having a corresponding position to the source pseudo-pixel is found among all the reference pseudo-pixels. That is, the reference pseudo-pixel located at the pixel position of (x, y) is found in the second pseudo-image.

Next, in step S430, according to the color and brightness difference between the source pseudo-pixel and the reference pseudo-pixel, it is determined whether the 2×2 pixel block corresponding to the source pseudo-pixel is to be time-stacked. This step is to prevent the first-order noise suppression image from generating ghosting when the color-to-luminance difference between the source pseudo-pixel and the reference quasi-pixel is too large.

In detail, first, the source-like blue chromaticity CBs, the source-like red chrominance CRs, the reference pseudo-blue chromaticity CBr, and the reference pseudo-red chrominance CRr are calculated. Suppose that Ys represents the pixel value of the source pseudo-pixel, Bs represents the pixel value of the corresponding blue channel in the 2×2 pixel block corresponding to the source pseudo-pixel, and Rs represents the 2×2 pixel block corresponding to the source pseudo-pixel. In this embodiment, the absolute difference between Ys and Bs is defined as the source pseudo-blue chromaticity CBs, and the absolute difference between Ys and Rs is defined as the source pseudo-red chrominance CRs. In addition, if Yr represents the pixel value of the reference pseudo pixel, Br represents the pixel value of the corresponding blue channel in the 2×2 pixel block corresponding to the reference pseudo pixel, and Rr represents the 2×2 pixel corresponding to the reference pseudo pixel. The pixel value corresponding to the red channel in the block, this embodiment defines the absolute difference between Yr and Br as the reference pseudo-blue chrominance CBr, and defines the absolute difference between Yr and Rr as the reference pseudo-red chrominance CRr.

Then, the luminance threshold value, the blue chrominance threshold value, and the red chrominance threshold value of the corresponding source pseudo-pixel are obtained. It must be specifically stated that the above threshold values are not fixed values, but will vary with the image features and brightness of the source-like pixels currently being processed. How to obtain the applicable brightness threshold, blue chrominance threshold and red chromatic threshold will be explained later with the icon.

When the absolute difference between the pixel value Ys of the source pseudo-pixel and the pixel value Yr of the reference pseudo-pixel is smaller than the luminance threshold, the absolute difference between the source pseudo-blue chrominance CBs and the reference pseudo-blue chrominance CBr is smaller than the blue chrominance threshold, and When the absolute difference between the source-like red chrominance CRs and the reference pseudo-red chrominance CRr is less than the red chromaticity threshold, it indicates that the color and brightness difference between the source pseudo-pixel and the reference pseudo-pixel is within the allowable range. It is determined that the pixel blocks corresponding to the source pseudo-pixels are time-stacked.

It is not difficult to imagine that in order to generate a better first-order noise suppression image, it is particularly important to determine whether or not to perform time-stacking processing on the pixel block. The value of the threshold value of the luminance threshold, the threshold of the blue chromaticity threshold, and the threshold value of the red chromaticity threshold are important parameters that affect the accuracy of the above-mentioned judgment mechanism.

Described below is a way to approximate a pixel of the current processing to obtain a corresponding luminance threshold. First, in the first pseudo-image, a pixel block of size m×m is obtained centering on the source pseudo-pixel, where m is a positive integer. Then, the average absolute error (MAE) of each pixel value in the m×m pixel block is calculated. Taking m as an example, FIG. 5 is a schematic diagram of a pixel block centered on a pixel (x, y) and having a size of 3×3. In this example, the average absolute error MAE is calculated as follows:

Where N is the number of pixels included in the m×m pixel block (in the embodiment shown in FIG. 5, N is 9), and Ys _i represents the pixel value of each pixel in the m×m pixel block, and Represents the pixel average of all pixels in an m×m pixel block. This embodiment will obtain the corresponding initial brightness value based on the average absolute error.

FIG. 6 is a graph showing the relationship between the average absolute error and the initial brightness value according to an embodiment of the invention. As shown in FIG. 6, the sizes of the first preset value a and the second preset value b are predefined and related to, for example, image features. When the value of the average absolute error is small (for example, less than M1), it indicates that the corresponding source pseudo-pixel system is in a relatively flat region of the image. Generally, the flat region in the image is more favorable for removing noise, so when the average absolute error is smaller than At M1, the corresponding initial brightness value is a second predetermined value b with a higher value, which can improve the chance of performing the time overlapping process. Conversely, when the average absolute error is larger (eg, greater than M2), it indicates that the source pseudo-pixel may have image features such as hard edges. Since the ghosting situation is easily generated by improperly superimposing the hard edges, when the average absolute error is greater than M2, the corresponding initial brightness value is the first preset value a with a lower value, thereby reducing the time stack. The opportunity to deal with. The average absolute error between M1 and M2 linearly corresponds to the initial luminance value between the first preset value a and the second preset value b.

FIG. 7 is a graph showing pixel values of source pseudo-pixels and luminance gain values according to an embodiment of the invention. As shown in FIG. 7, the preset third preset value c and the fourth preset value d may be related to image brightness, for example. The smaller the pixel value of the source pseudo-pixel (for example, less than Y1), the lower the brightness. Since the dark part of the image is more likely to have more noise, in order to eliminate the noise as much as possible, when the pixel value of the source pseudo pixel is less than Y1, the corresponding brightness gain value will be the fourth preset value with higher value. d, this will increase the chances of time stacking. Conversely, when the pixel value of the source pseudo-pixel is larger (for example, greater than Y2), it indicates that it has higher brightness, so if the pixel value of the source pseudo-pixel is greater than Y2, the corresponding brightness gain value will be lower. The third preset value c is to reduce the chance of performing the time superposition process. In the case where the pixel value of the source pseudo pixel is between Y1 and Y2, the corresponding brightness gain value is a value between the third preset value c and the fourth preset value d.

After obtaining the luminance initial value and the luminance gain value respectively, the present embodiment takes the product of the luminance initial value and the luminance gain value as the luminance threshold value of the corresponding source pseudo-pixel.

In this embodiment, a blue chrominance gain value is obtained according to the calculated source pseudo blue chromaticity CBs in a manner similar to the above, and the product of the luminance initial value corresponding to the average absolute error and the blue chrominance gain value is taken as the blue chromaticity. Threshold value. The red chrominance threshold is obtained in a similar manner to the blue chrominance threshold. After obtaining a red chrominance gain value based on the source pseudo-red chrominance CRs, the initial luminance value and the red chromaticity corresponding to the average absolute error are obtained. The product of the gain values is taken as the red chrominance threshold.

Please return to FIG. 4, if the result of the determination in step S430 is that the time superimposing process is not performed, then as shown in step S440, the 2×2 pixel block corresponding to the source pseudo-pixel is directly used as the partial superposition of the corresponding source pseudo-pixel. result.

If the time superimposing process is to be performed, as shown in step S450, each pixel value in the pixel block of the size 2×2 corresponding to the source pseudo-pixel is compared with the first dynamic compensation result, the second dynamic compensation result, The third dynamic compensation result, and the pixel values at the corresponding positions in the fourth dynamic compensation result are superimposed, thereby generating a partial overlapping result corresponding to the source pseudo-pixel. In one embodiment, the method of superimposing is to calculate an average of two pixel values corresponding to the position. In another embodiment, the time overlap weight value corresponding to the currently processed source pseudo-pixel is obtained, and the partial overlap result is generated according to the time overlap weight value.

FIG. 8 is a graph showing the relationship between the average absolute error and the time overlap weight value according to an embodiment of the invention. As shown in FIG. 8, when the average absolute error is larger (for example, greater than M4), it means that the source pseudo-pixel may have image features such as hard edges, in order to generate a first-order noise suppression image with clear edges, obtained in the foregoing manner. After the average absolute error, if the average absolute error is greater than M4, the corresponding time overlap weight value is the second weight preset value f with a higher value. If the average absolute error is less than M3, the corresponding time overlap weight value is the first weight preset value e with a lower value. If the average absolute error is between M3 and M4, it will linearly correspond to the time overlap weight between the first weight preset value e and the second weight preset value f.

In this embodiment, for example, the partial superposition result BR of the corresponding source pseudo-pixel is calculated by the formula (1):

Wherein, Wc represents a time overlap weight value, and C_base represents a preset coefficient (for example, 128 or 512, which is not limited in the present invention). Rs, Grs, Gbs, and Bs respectively represent pixel values of the corresponding red channel, the first green channel, the second green channel, and the blue channel in the 2×2 pixel block corresponding to the source pseudo pixel. And Ra, Gra, Gba, and Ba respectively indicate that in the first dynamic compensation result, the second dynamic compensation result, the third dynamic compensation result, and the fourth dynamic compensation result, in the 2×2 pixel block corresponding to the source pseudo pixel Each pixel is at the pixel value of the corresponding position.

Taking the first pseudo-image 300 of FIG. 3 as an example, it is assumed that the source pseudo-pixel 210_b in the first pseudo-image 300 is currently processed. In the case where it is determined that the pixel block 210_a corresponding to the source pseudo-pixel 210_b is to be time-stacked, the upper left pixel (which has the pixel value corresponding to the first green channel) in the pixel block 210_a is to be superimposed. If the pixel position of the upper left pixel in the source image 200 is (1, 1) and the pixel value is Grs, this embodiment will find that the second dynamic compensation result corresponding to the first green channel belongs to (1, 1) The pixel value Gra of this pixel position, and then Grs and Gra are substituted into the equation (Wc × Grs + (C_base - Wc) × Gra) / C_base to find the superposition result of the upper left pixel of the pixel block 210_a. The result of superposition of the upper right, lower left, and lower right pixels of the pixel block 210_a can also be obtained in a similar manner. The result of these four superpositions is the partial superposition result corresponding to the source pseudo-pixel 210_b.

In this embodiment, step S420 to step S470 are repeatedly performed to obtain corresponding partial overlapping results for each of the source pseudo-pixels. Finally, as shown in step S480, the partial superposition results corresponding to the respective pseudo-pixels of the respective sources are reconstructed according to the Belle color arrangement rule to generate the first-order noise suppression image. The first-order noise suppression image is the same size as the source image.

Taking the first pseudo image 300 of FIG. 3 as an example again, if (α, β) is used, the pixel position in the corresponding first-order noise suppression image is represented, where α is between 1 and 8 and β is between 1 and Between 6. If (1,1) is the pixel position of the top left corner of the first-order noise suppression image, and (8,6) is the pixel position of the bottom right corner of the first-order noise suppression image, then the source is similar to the pixel 210_b. The partial superposition result will be the pixel values of the four pixels located at (1,1), (1,2), (2,1), and (2,2) according to the Belle color arrangement rule, and the source is pseudo-like The partial superposition result corresponding to the pixel 220_b will be respectively used as the pixels of the four pixels (1, 3), (1, 4), (2, 3), and (2, 4) according to the Belle color arrangement rule. Value, and so on.

For a digital image capturing device that continuously captures multiple images, the second-order noise suppression image generated by using each step shown in FIG. 1 can be used as a noise for removing the next source image. Reference image. In detail, after generating the first second-order noise suppression image based on the first and second source images, the first second-order noise can be obtained when preparing to remove the noise of the third source image. The image is suppressed as a reference image to obtain a second second-order noise suppression image. The second second-order noise suppression image can also be used as a reference image when the fourth source image is subjected to noise removal processing. In this way, cascaded three-dimensional noise removal can be achieved in a limited hardware resource environment.

It should be particularly noted that although the present invention is described by taking a reference image as an example in the above embodiment, in other embodiments, multiple reference images may be used to remove noise in the source image. When a plurality of reference images are used, if it is determined that the partial superposition result of a source pseudo-pixel is to be calculated by the above formula (1), the weight value of the portion of (C_base-Wc) is equally divided into the respective reference images. Since the steps of generating the second-order noise suppression image by using the plurality of reference images are the same as or similar to those of the foregoing embodiment, they are not described herein again.

In summary, the noise suppression method of the present invention utilizes the correlation between the source image and the reference image, analyzes the color information, and performs time superposition processing and spatial denoising processing to achieve three-dimensional noise removal. effect. Among them, the details of the image and the degree of lightness and darkness are taken into consideration when performing the time superimposition processing, thereby avoiding overlapping images with ghosting conditions. Space-to-noise processing further eliminates noise, which results in better noise removal results.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the invention, and any one of ordinary skill in the art can make some modifications and refinements without departing from the spirit and scope of the invention. The scope of the invention is defined by the scope of the appended claims.

S110~S150‧‧‧ steps of the noise suppression method according to an embodiment of the present invention

200‧‧‧ source image

210_a, 220_a‧‧‧ pixel blocks

300‧‧‧First quasi-image

210_b, 220_b‧‧‧ source quasi-pixel

S410~S480‧‧‧ steps of generating a first-order noise suppression image according to an embodiment of the present invention

A‧‧‧first preset value

B‧‧‧second preset value

M1, M2, M3, M4‧‧‧ average absolute error

C‧‧‧ third preset value

D‧‧‧ fourth preset

Y1, Y2‧‧‧ source pixel-like pixel values

e‧‧‧First weight preset

F‧‧‧second weight preset

FIG. 1 is a flow chart of a method for suppressing noise according to an embodiment of the invention.

2 is a schematic diagram of a source image according to an embodiment of the invention.

3 is a schematic diagram of a first pseudo-image as depicted in an embodiment of the invention.

4 is a flow chart of generating a first-order noise suppression image according to an embodiment of the invention.

FIG. 5 is a schematic diagram of a pixel block centered on a source pseudo-pixel according to an embodiment of the invention.

FIG. 6 is a graph showing the relationship between the average absolute error and the initial brightness value according to an embodiment of the invention.

FIG. 7 is a graph showing pixel values of source pseudo-pixels and luminance gain values according to an embodiment of the invention.

FIG. 8 is a graph showing the relationship between the average absolute error and the time overlap weight value according to an embodiment of the invention.

S110～S150. . . Each step of the noise suppression method according to an embodiment of the present invention

Claims (5)

A noise suppression method for filtering noise in a source image by using a reference image, wherein the source image and the reference image respectively comprise n 2×2 pixel blocks, and each of the 2×2 pixel blocks Each of the pixels has a corresponding R color channel, a first green channel (Gr color channel), a second green channel (Gb color channel) according to a Bayer pattern color arrangement rule. a pixel value of one of the B color channels, and n is a positive integer, the method comprising: generating a first pseudo image corresponding to the source image and a second pseudo image corresponding to the image, wherein The first pseudo-image includes n source pseudo-pixels, and each of the source pseudo-pixels respectively corresponds to one of the 2×2 pixel blocks in the source image, and the second quasi-image includes n reference quasi-pixels And each of the reference pseudo-pixels respectively corresponds to one of the 2×2 pixel blocks in the image; calculating a global motion vector between the first quasi-image and the second quasi-image; The global motion vector performs a dynamic compensation process on the reference image to obtain a plurality of dynamic compensation results. The dynamic compensation result is used to perform a temporal blending process on the source image to generate a first-order noise suppression image. Included, for each of the sources, a pseudo-pixel, in which the reference quasi-like image has a corresponding position corresponding to the source pseudo-pixel And determining, according to the color and brightness difference between the source pseudo-pixel and the reference pseudo-pixel, whether to perform the time superimposition processing on the 2×2 pixel block corresponding to the source pseudo-pixel, including: the source pseudo-pixel The absolute difference between the pixel value and the pixel value corresponding to the blue channel in the 2×2 pixel block corresponding to the source pseudo-pixel is defined as a source pseudo-blue chromaticity; the pixel value of the source pseudo-pixel and the source The absolute difference of the pixel values corresponding to the red channel in the 2×2 pixel block corresponding to the pseudo-pixel is defined as a source pseudo-red chromaticity; the pixel value of the reference pseudo-pixel corresponds to the reference pseudo-pixel 2× The absolute difference value of the pixel value corresponding to the blue channel in the 2-pixel block is defined as a reference pseudo-blue chromaticity; the pixel value of the reference pseudo-pixel is in the 2×2 pixel block corresponding to the reference pseudo-pixel The absolute difference of the pixel values of the red channel should be defined as a reference pseudo-red chromaticity; a luminance threshold corresponding to the source-like pixel, a blue chrominance threshold, and a red The chrominance threshold includes: obtaining, in the first pseudo-image, an m×m pixel block centered on the source pseudo-pixel, wherein m is a positive integer; calculating each pixel value in the m×m pixel block Mean Absolute Error (MAE); obtaining an initial brightness value based on the average absolute error; Obtaining a brightness gain value according to the pixel value of the source pseudo pixel; taking the product of the brightness initial value and the brightness gain value as the brightness threshold corresponding to the pseudo pixel; obtaining a blue color according to the source pseudo blue color a gain value; a product of the luminance initial value and the blue chrominance gain value as a blue chrominance threshold corresponding to the source pseudo pixel; a red chrominance gain value obtained from the source pseudo red chromaticity; and the luminance The product of the initial value and the red chrominance gain value is used as the red chrominance threshold corresponding to the source pseudo pixel; and the absolute difference between the pixel value of the source pseudo pixel and the pixel value of the reference pseudo pixel is less than the brightness threshold a value, the absolute difference between the source-like blue chromaticity and the reference pseudo-blue chromaticity is less than the blue chromaticity threshold, and the absolute difference between the source-like red chromaticity and the reference pseudo-red chromaticity is less than the red chromaticity When the threshold is thresholded, it is determined that the time overlap processing is performed on the 2×2 pixel block corresponding to the source pseudo pixel; and the first order is The noise suppression image performs a spatial noise reduction process to generate a second order noise suppression image. The method for suppressing noise according to claim 1, wherein the step of performing the dynamic compensation processing on the reference image according to the global motion vector to obtain the dynamic compensation results comprises: Aligning the source image according to the global motion vector to generate a first dynamic compensation result corresponding to the red channel, the first green channel, the second green channel, and the blue channel, and a second dynamic compensation The result, a third dynamic compensation result, and a fourth dynamic compensation result. The method for suppressing noise according to claim 2, wherein the step of performing the time superimposition processing on the source image by using the dynamic compensation results to generate the first-order noise suppression image further comprises: when determining When the source is similar to the 2×2 pixel block corresponding to the pixel, when the time overlap processing is performed, each pixel value in the 2×2 pixel block corresponding to the source pseudo pixel is compared with the first dynamic compensation result. The second dynamic compensation result, the third dynamic compensation result, and the pixel values at the corresponding positions in the fourth dynamic compensation result are superimposed to generate a partial overlapping result corresponding to the source pseudo pixel; when the source is not determined When the 2×2 pixel block corresponding to the pseudo-pixel is subjected to the time superimposition processing, the 2×2 pixel block corresponding to the source pseudo-pixel is used as the local superposition result corresponding to the pseudo-pixel; and according to Bell The color arrangement rule reorganizes the partial superposition results corresponding to the respective pseudo-pixels of the source to generate the first-order noise suppression image. The method for suppressing noise according to claim 3, wherein the step of generating the partial overlap result corresponding to the source of the pseudo-pixels further comprises: And obtaining a time overlap weight value according to the average absolute error; and generating the partial overlap result corresponding to the source pseudo pixel according to the time overlap weight value. The method for suppressing noise according to claim 4, wherein the step of generating the partial overlap result corresponding to the source pseudo pixel according to the time overlap weight value comprises calculating by the formula (1): Where BR denotes the partial superposition result corresponding to the source pseudo-pixel, Wc denotes the time superimposed weight value, C_base denotes a preset coefficient, and Rs, Grs, Gbs, Bs respectively represent 2× corresponding to the source pseudo-pixel The pixel values corresponding to the red channel, the first green channel, the second green channel, and the blue channel in the 2 pixel block, Ra, Gra, Gba, and Ba respectively represent the first dynamic compensation result and the second In the dynamic compensation result, the third dynamic compensation result, and the fourth dynamic compensation result, each pixel in the 2×2 pixel block corresponding to the source pseudo pixel is at a pixel value corresponding to the position.