TW201408059A - Image noise filtering method and device - Google Patents

Abstract

An image noise filtering method is provided, which is suitable for filtering out the noise of an input image so as to obtain an output image. At first, the difference between an image pixel I(x, y) and I(x-(p-1)/2, y-(p-1)/2) is calculated to be used as a deviation element. Subsequently, those deviation elements corresponding to an comparison block of the pixel I(x, y) are compared with those deviation elements corresponding to a comparison block of the adjacent pixel so as to determine a difference element of the pixel I(x, y). Afterwards, the difference elements of the adjacent pixels are utilized to constitute a summation of all deviation elements corresponding to the comparison block of the pixel I(x, y), which is used as a block element. Finally, based on the block element of each pixel I(x, y) and the pixel I(x-(p-1)/2, y-(p-1)/2), a weighing operation is performed to obtain the pixel O(x, y) of the output image.

Description

Image noise filtering method and device

The invention relates to an image processing method, in particular to an image noise filtering method.

During the process of capturing, converting or transmitting images, some undesired noise interference is inevitably introduced. Therefore, the interference image is usually filtered to improve the picture quality, and then the subsequent monitoring is performed. System image analysis or other image optimization processing.

Referring to FIG. 1, a conventional noise filtering method uses a non-local average (Non-Local Means (NLM)) to configure a filter with a p×p pixel centered on a pixel to be processed in an image. Window, and configure a pixel (centered as filtered window pixel) in the filtering window with a matching block centered on it and having q×q pixels, q<p. Then, the comparison block of the pixel to be processed is compared with the comparison block of each filter window pixel to calculate an output pixel corresponding to the pixel to be processed after the noise filtering.

The calculation method for obtaining the output pixel is: calculating the sum of the pixel differences of the comparison block of the pixel to be processed and the comparison block of one of the filtered window pixels at the corresponding position, and then calculating the to-be-processed picture The sum of the pixel differences between the aligned blocks of the prime and the other block of the filtered window pixels at the corresponding positions until the sum of the pixel differences for all filtered window pixels is calculated, and then the calculated regions are used The sum of pixel differences between blocks, weighted average of the filtered window pixels to obtain the output pixels. However, there are many overlapping blocks of adjacent pixels, so there are many duplicates in the above methods. Counting, resulting in inefficient computing, wastes computational costs in vain.

Therefore, the object of the present invention is to provide an image noise filtering method and apparatus, which can effectively reduce the calculation cost.

Therefore, the image noise filtering method of the present invention is suitable for filtering out noise of an input image to obtain an output image, wherein the H×W pixels of the input image form an H×W input element I of an input array ( x , y ), the H×W pixels of the output image form an H×W output elements O( x , y ) of an output array, 0≦x<W, 0≦y<H, the image noise filter The dividing method comprises the following steps: (A) assembling an offset array generator, calculating an offset element for each input element I( x , y ), and the offset element corresponding to each input element I( x , y ) represents The input element I( x , y ) and another input element The difference, 0<p<W, 0<p<H, p is a positive odd number; (B) is combined with a difference generating unit for each input element I( x , y ), making one of the input elements aligned Those offset elements corresponding to the block are compared to those corresponding to one of the adjacent input elements, and a difference element is obtained, wherein the comparison element of the input element includes the input element And a plurality of adjacent input elements; (C) a computing unit, for each input element I( x , y ), a block element is calculated according to the difference element, wherein the block element represents based on the input element I( x , y ) is the sum of all the offset elements corresponding to the block; and (D) is an output unit for each input element I( x , y ), based on the corresponding block element and the other Input element , the output element O( x , y ) corresponding to the input element I( x , y ) is calculated.

The image noise filtering device of the present invention is adapted to filter out noise of an input image to obtain an output image, wherein the H×W pixels of the input image form an H×W input element I of the input array ( x) , y ), the H×W pixels of the output image form an H×W output element O( x , y ) of an output array, 0≦x<W, 0≦y<H, the image noise filtering The apparatus includes: an offset array generator that calculates an offset element for each input element I( x , y ), and an offset element corresponding to each input element I( x , y ) represents the input element I( x , y And another input element The difference, 0<p<W, 0<p<H, p is a positive odd number; a difference generating unit, for each input element I( x , y ), such that one of the input elements corresponds to the corresponding block The offset element obtains a difference element compared to the offset elements corresponding to one of the adjacent input elements, wherein the comparison element of the input element includes the input element and the adjacent plurality of An input element; a calculation unit, for each input element I( x , y ), calculates a block element according to the difference element, wherein the block element represents an alignment area based on the input element I( x , y ) The sum of all offset elements corresponding to the block; and an output unit for each input element I( x , y ) based on the corresponding block element and the other input element , the output element O( x , y ) corresponding to the input element I( x , y ) is calculated.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments of the invention.

Before the present invention is described in detail, it is noted that in the following description, similar elements are denoted by the same reference numerals.

First preferred embodiment

Referring to FIG. 2 and FIG. 3, the first preferred embodiment of the image noise filtering device 100 of the present invention is adapted to filter out noise of an input image to obtain an output image. The H×W pixels of the input image form an input array with H×W input elements I( x , y ), and the H×W pixels of the output image form one H×W output elements O( x , The output array of y ), and the different array elements hereinafter referred to as the same array location, correspond to each other.

The image noise filtering device 100 configures each input element I( x , y ) with a filtering window centered on it and having p×p input elements, and configures one input element in the filtering window as the upper left corner. Alignment block, wherein the comparison block has r × r input elements, 0 < r < p < W, 0 < r < p < H, r and p are both positive odd numbers, 0 ≦ x < W, 0 ≦y<H. Note that for the convenience of the diagram, Figure 2 only depicts the input elements I( x , y ) and Aligned blocks, but in fact each input element in the filter window is configured with a matching block.

The image noise filtering device 100 operates on the principle that the comparison block of the input element to be processed is compared with the comparison block of each input element in the filtering window, and the input elements in the filtering window are compared according to each other. The result is weighted averaged to get the corresponding output element. However, as for the input elements and adjacent input elements to be processed, the filtering windows of the two overlap and there are more overlaps between the blocks, so the calculation process of the corresponding output elements is extremely repetitive. The calculation efficiency is not good. Therefore, in this embodiment, the comparison result of the comparison block of the input element to be processed and the comparison block of each input element in the filtering window is gradually pieced together according to the difference of the adjacent comparison block, and the output is obtained. element.

In detail, the image noise filtering device 100 of FIG. 3 includes a setting unit 1 and a receiving unit 2 and an offset array sequentially connected in sequence. The unit 3, a block calculator 4, a weight processor 5 and an output unit 6.

The receiving unit 2 receives the input array having H columns and W rows. The setting unit 1 determines the number of input elements of the filter window (i.e., p × p) and the number of input elements of the aligned block (i.e., r × r). The offset array generator 3 configures a filter element for each input element I( x , y ) of the input array And obtaining an offset array of the same array size according to the input array, wherein each offset array element D( x , y ) (hereinafter referred to as an offset element) is a corresponding input element I( x , y ) and a filter element The difference. The block calculator 4 then finds a block array of array size 1×W according to the offset array, and updates the block array for each column of the offset array, wherein the block array element N ( x ) updated for each column (hereinafter referred to as a block element) is equivalent to the sum of all corresponding offset elements in the aligned block of the corresponding input element I( x , y ) in the column. Next, the weight processor 5 uses the block elements to find a weight array of array size H×W, and uses the corresponding block elements and filtering elements to find a weighted array of array size H×W. Finally, the output unit 6 performs a division operation based on the weight array and the weighted array to obtain the output elements of the output array.

More specifically, the block calculator 4 of the present embodiment includes a difference generating unit 41 and a calculating unit 42 in order to effectively reduce the computational cost of the obtained block array. The difference generating unit 41 finds a difference array having H × W difference elements M( x , y ) based on the offset array, and then supplies the calculation unit 42 to calculate the block elements corresponding to the input elements I( x , y ) of the respective columns. N( x ).

Referring to FIG. 4, the steps involved in the preferred embodiment of the image noise filtering apparatus 100 for performing the image noise filtering method of the present invention will be described next.

Step 70: The setting unit 1 determines the number of input elements of the filtering window p× p, and determine the number of input elements r × r of the comparison block.

Step 71: The receiving unit 2 receives the input element I( x , y ) of the input array, O≦x<W, 0≦y<H.

Step 72: Offset Array Generator 3 configures a filter element for each input element I( x , y ) And calculate the elements I( x , y ) and Difference to obtain the input element I (x, y) corresponding to the deviation element D (x, y).

Please note that when or The offset element D( x , y )=0.

Step 73: For each input element I( x , y ), the difference generating unit 41 uses a ratio corresponding to the input elements I( x , y ), I( x -1, y ), I( x , y -1) For those offset elements of the block, the difference element M( x , y ) is found. Since the difference element M( x , y ) for processing the image boundary needs to consider the boundary problem, step 73 is divided into the following sub-steps:

Sub-step 731: The difference generating unit 41 calculates the first column difference element M( x , 0).

That is to say, when x=0, all the corresponding offset elements D( x + △ x , y + Δ y ) in the aligned block of the input elements I( x , y ) are added to obtain the difference element M ( x , 0). When 0 < x ≦ r, only r D ( x + r , y + Δ y ) are added. When r < x ≦ (W-1-r), in addition to adding r D ( x + r , y + Δ y ), r D ( x -1, y + Δ y ) are deducted. When x>(W-1-r), only r D ( x -1, y + Δ y ) are deducted. In short, when processing the first column difference element, only in the case of x=0, the sum of all corresponding offset elements D( x + △ x , y + Δ y ) in the comparison block is completely obtained. For M( x , 0), the other case is to calculate the difference between the "currently the sum of the block offset elements" compared to the "I( x -1, y ) comparison block offset element sum". Refer to Figure 5. Moreover, in order to facilitate the illustration, FIG. 5 takes r=5 as an example, but the actual application is not limited thereto, and the following FIGS. 6-8 also simplifies the diagram with r=5.

Sub-step 732: The difference generating unit 41 calculates the yth column difference element M( x , y ), 0<y≦r.

When processing the y (0 < y ≦ r) column difference element, only in the case of x = 0, add r D ( x + △ x , y + r ) as M ( x , y ), other The case is to calculate the difference between "the sum of the current comparison block offset elements" as compared with the "I( x -1, y ) comparison block offset element sum" as x increases, which can be referred to FIG.

Sub-step 733: The difference generating unit 41 calculates the yth column difference element M( x , y ), r<y≦(Hr-1).

When processing the y (r<y≦(Hr-1)) column difference element, only in the case of x=0, add r D ( x + △ x , y + r ) and deduct r D ( x + △ x , y -1) are treated as M( x , y ), and in other cases, the sum of the current matching block offset elements is calculated as compared with the "adjacent comparison block offset element sum" The difference between the adjacent alignment blocks referred to herein is the alignment block of I( x -1, y ), I( x , y -1), which can be referred to FIG.

Sub-step 734: The difference generating unit 41 calculates the yth column difference element M( x , y ), (Hr-1) < y ≦ (H-1).

When processing the difference element of the y((Hr-1)<y≦(H-1)) column, only r ( X + △ x , y -1) is deducted as x = 0 M( x , y ), in other cases, the difference between the current sum of the block offset elements and the sum of the adjacent block offset elements. The adjacent block is called here. It refers to the alignment block of I( x -1, y ), I( x , y -1), which can be referred to Figure 8.

Step 74: Next, the flow calculates a corresponding weight element W( x , y ) and a weighting element S( x , y ) for each input element I( x , y ) column by column, wherein the weight element W( x , y) ) is an element of the weight array, and the weighting elements S( x , y ) are elements of the weighted array.

When processing each column, the process has the following four substeps.

Sub-step 740: The computing unit 42 first causes the temporary storage elements N_tmp(0), N_tmp(1)...N_tmp( W -1) to be 0.

Sub-step 741: The calculation unit 42 is the current input element I( x , y ) of the column (assumed to be the yth column), and uses the corresponding difference element M( x , y ) and the temporary storage element N_tmp( x ) to find the block. The element N( x )= N _ tmp ( x )+M(x,y), and the temporary element N_tmp( x )= N (x). That is, the yth column block element .

Sub-step 742: The weight processor 5 updates the current pre-signal LW=exp(- N ( x )/( p ² ) for the current input element I( x , y ) of the column using the corresponding block element N( x ) σ )), and the weight element W( x , y ) is set as in equation (5), where σ is an adjustment parameter that can control the filtering effect. In other words, the y column weight element .

Sub-step 743: The weight processor 5 is the current input element I( x , y ) of the column, using the corresponding weight preamble and filtering elements Update a weighted preamble And set the weighting element S( x , y ) as in equation (6). In other words, the y column weighting element .

Then, sub-steps 740-743 are repeated to calculate the weight elements and weighting elements of the next column of input elements. Until all columns have been processed, the weight processor 5 can take all the elements of the weighted array and the weighted array.

Step 75: The output unit 6 divides the weighting elements corresponding to the input elements I(x, y) by the weight elements to obtain the output elements O(x, y), and aggregates all the output elements O(x, y) to form a noise filter. The output image after the removal.

In summary, the calculation principle of step 73 for the yth column difference element M( x , y ) is as follows. In the case of y=0, when x=0, the difference element represents the sum of all the offset elements corresponding to the block of the input element I( x , y ); when 0<x<W, the difference element represents "The sum of the input elements I( x , y ) corresponds to the sum of all offset elements of the block" compared to the sum of the input elements I( x -1, y ) for all offset elements corresponding to the block The difference.

On the other hand, in the case of 0 < y ≦ r, when x = 0, the difference element represents "the sum of all the offset elements corresponding to the block of the input element I ( x , y )" compared to " The difference between the input element I( x , y -1) and the sum of all offset elements corresponding to the block; when 0<x<W, the difference element represents the ratio of the input element I( x , y ) The sum of all the offset elements corresponding to the block is compared to the sum of the ratios of the input elements I( x -1, y ) to the total of all the offset elements corresponding to the block.

Moreover, in the case of r<y<H, when x=0, the difference element represents “the sum of all the offset elements corresponding to the block of the input element I( x , y )” compared to the “input” The difference between the ratio of the elements I( x , y -1) to the sum of all offset elements corresponding to the block; when 0<x<W, the difference element represents the comparison region of the input element I( x , y ) The sum of all offset elements corresponding to the block is compared to the sum of the ratios of the input elements I( x -1, y ) and I( x , y -1) to all offset elements corresponding to the block.

Furthermore, in step 74, the weight preamble signal LW = exp(- N ( x ) / ( p ² σ )), which implies that the larger block element N( x ) will correspond to a smaller LW. The preferred value of the adjustment parameter σ is 5.5, and the larger the σ, the better the filtering effect.

Second preferred embodiment

Compared with the first preferred embodiment, the second preferred embodiment of the image noise filtering device 100 of the present invention differs in the following three points.

(1) The offset array size H×W of the first embodiment, and the second embodiment calculates the most reference to the (y-1)~(y+r-1) for the difference element of the yth column in the sub-steps 731-734. The column offset element, so only the (r+1) column offset element is stored as the 0th to the rth columns of the offset array, that is, the offset array size (r+1)×W.

(b) The difference array size H×W of the first embodiment, and the second embodiment only stores the nearest one column difference element as the difference array because the sub-step 741 calculates only the reference element of the yth column for the block element calculation. That is, the difference array size is 1×W.

(3) The weight array and the weighting array of the first embodiment each have an array size H×W, and the second embodiment adjusts the flow of steps, storing only the nearest one column weight element as a weight array, and storing only the most recent one. The column weighting elements are treated as a weighted array, that is, the array size of the weighted array and the weighted array are both 1×W.

In detail, the steps of the second embodiment adjustment are as follows.

Step 72': The offset element D'( x , i )= D ( x , y + i -1), 0≦i≦r obtained by the offset array generator 3.

Step 73': the difference generating unit 41 is the current input element I( x , y ) of the yth column, and finds the difference element M using those offset elements corresponding to the comparison block of the input element I( x , y ) '( x ), where y=0 is based on equation (7), 0<y≦r is based on equation (8), and r<y≦(Hr-1) is according to equation (9), (Hr- 1) <y≦(H-1) is based on equation (10).

Sub-step 740': The calculation unit 42 causes the temporary elements N_tmp(0), N_tmp(1)...N_tmp( W -1) to be 0.

Sub-step 741': the calculation unit 42 is the current input element I( x , y ) of the column, and uses the corresponding difference element M'( x ) and the temporary storage element N_tmp( x ) to find the block element N( x )= N _ tmp ( x )+M'(x) and let the scratch element N_tmp( x )= N (x).

Sub-step 742': The weight processor 5 is the current input element I( x , y ) of the column, and updates the weight preamble signal LW=exp(- N ( x )/( p ² ) using the corresponding block element N( x ) σ )), and the weight element W'( x ) is set as in equation (7), where σ is an adjustment parameter that controls the filtering effect.

Sub-step 743': weight processor 5 is the current input element I( x , y ) of the column, using the corresponding weight preamble and filtering elements Update weighted preamble And set the weighting element S'( x ) as in equation (8).

Next, step 75' is performed: the output unit 6 divides the weighting element corresponding to each input element I(x, y) of the column by the weight element to obtain the output element O(x, y) of the yth column.

Then, sub-steps 73', 740', 741', 742', 743' and step 75' are performed again to calculate the output element O(x, y) of the (y+1)th column. Until all columns are processed, all output elements O(x, y) can be aggregated to form noise filtering. After the output image.

It should be reminded that the foregoing process is performed in a column-by-column manner, but those skilled in the art can easily infer how to calculate the weighting elements and weighting elements line by line to obtain the output image for filtering noise, so no more is added here. Description.

Furthermore, although the present embodiment illustrates that both the filtering window and the comparison block are square, in other applications, the filtering window and the comparison block may also be rectangular or other, as long as the comparison block <filter window.

In addition, although the embodiment illustrates that the filtering window is centered on the corresponding input element, the comparison block is the upper left corner of the corresponding input element, but in other applications, the corresponding input element is in the filtering window and the position of the comparison block. It is not limited to the above, as long as the filter window is composed of adjacent elements of the corresponding input elements, and the comparison block is composed of adjacent elements of the corresponding input elements. It should be noted that if the input element is located at another specific location of the aligned block, the filter element position must be adjusted according to the block size. For example, when the input element I( x , y ) is in the lower right corner of the comparison block, you can choose Used as a filter element.

In summary, in the foregoing preferred embodiment, the block calculator 4 uses the difference generating unit 41 to find the difference between the sum of the offset elements D(x, y) in the adjacent matching block, and then combines using the calculating unit 42. The sum of the offset elements D(x, y) in each of the aligned blocks, that is, the block elements, is further calculated by the weight processor 5 to obtain the weight elements and the weighting elements in an accumulated manner, so that the output unit 6 obtains the output image. The calculation cost is significantly less than the conventional technology, so the object of the present invention can be achieved.

The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent.

100‧‧‧Image noise filtering device

1‧‧‧Setting unit

2‧‧‧ Receiving unit

3‧‧‧Offset Array Generator

4‧‧‧block calculator

41‧‧‧balance generating unit

42‧‧‧Computation unit

5‧‧‧weight processor

6‧‧‧Output unit

70~75‧‧‧Steps

1 is a schematic diagram illustrating a filtering window and a comparison block configured by an input technique for an input element; FIG. 2 is a schematic diagram illustrating a filtering window and a comparison block configured for an input element in the preferred embodiment; FIG. Is a block diagram illustrating the image noise filtering device of the first preferred embodiment; FIG. 4 is a flow chart illustrating the image noise filtering method of the preferred embodiment; and FIGS. 5-8 are schematic diagrams illustrating Find the difference element based on the offset element.

100‧‧‧Image noise filtering device

1‧‧‧Setting unit

2‧‧‧ Receiving unit

3‧‧‧Offset Array Generator

4‧‧‧block calculator

41‧‧‧balance generating unit

42‧‧‧Computation unit

5‧‧‧weight processor

6‧‧‧Output unit

Claims (10)

An image noise filtering method is provided for filtering out noise of an input image to obtain an output image, wherein the H×W pixels of the input image form an H×W input element I ( x , y of an input array) The H×W pixels of the output image form an H×W output elements O( x , y ) of an output array, 0≦x<W, 0≦y<H, and the image noise filtering method includes The following steps: (A) grouping an offset array generator, calculating an offset element for each input element I( x , y ), and the offset element corresponding to each input element I( x , y ) represents the input element I( x , y ) and another input element Difference, 0<p<W, 0<p<H, p is a positive odd number, and when or When the offset element is 0; (B) is combined with a difference generating unit for each input element I( x , y ) such that one of the input elements is compared with those corresponding to the offset block One of the adjacent input elements aligns the offset elements corresponding to the block to obtain a difference element, wherein the comparison element of the input element includes the input element and a plurality of adjacent input elements; (C) Composing a calculation unit for each input element I( x , y ), and calculating a block element according to the difference element, wherein the block element represents a comparison block based on the input element I( x , y ) The sum of all offset elements corresponding to; and (D) an output unit for each input element I( x , y ) based on the corresponding block element and the other input element , the output element O( x , y ) corresponding to the input element I( x , y ) is calculated. According to the image noise filtering method of claim 1, wherein in step (B), in the case of y=0, when x=0, the difference element calculated by the difference generating unit represents the input The ratio of the elements I( x , y ) to the sum of all the offset elements corresponding to the block, when 0<x<W, the difference element calculated by the difference generating unit represents the comparison area of the input element I( x , y ) The sum of all offset elements corresponding to the block compared to the sum of the input elements I( x -1, y ) for the sum of all offset elements corresponding to the block; in the case of 0 < y ≦ r, when x = 0, the difference element calculated by the difference generating unit represents the ratio of the input element I( x , y ) to the sum of all the offset elements corresponding to the block compared to the input element I( x , y -1) The difference between the sum of all the offset elements corresponding to the block, when 0<x<W, the difference element calculated by the difference generating unit represents the sum of all the offset elements corresponding to the block of the input element I( x , y ) compared to the input element I (x -1, y) of the ratio of the difference of the sum of all blocks corresponding shift elements; in the case of r <y <H when x = 0, the The generating element represents the amount of difference calculating unit, the input element I (x, y) of the ratio of the sum of all elements of the block corresponding to the offset as compared to the input element I (x, y -1) corresponding to the ratio of the block The difference between the sum of all the offset elements, when 0 < x < W, the difference element calculated by the difference generating unit represents the ratio of the input elements I ( x , y ) to the sum of all the offset elements corresponding to the block The difference between the ratio of the input elements I( x -1, y ), I( x , y -1) to the sum of all the offset elements corresponding to the block. According to the image noise filtering method of claim 1, wherein in the step (B), the ratio of the input elements I( x , y ) used by the difference generating unit to the comparison block has r× r input elements, and the input elements I( x , y ) are those in the upper left corner of the r×r input elements, 0<r<p<W, 0<r<p<H. The image noise filtering method according to Item 3 of the patent application scope further comprises a step (E) between the step (C) and the step (D): assembling a weight processor for each input element I ( x , y ), using the corresponding block element N( x ) and the other input element Setting a weight element and a weighting element; and in step (D), the output unit divides the weighting element of each input element I( x , y ) by the weight element to obtain a corresponding output element O( x , y ): wherein, for the input element of the yth column, the weight processor sets the weight element of the input element I( x , y ) And σ is an adjustment parameter that can control the filtering effect, and r×r is the number of input elements of each comparison block. According to the image noise filtering method of claim 4, in the step (E), for the input element of the yth column, the weight processor sets the weight of the input element I( x , y ) element . An image noise filtering device is provided for filtering out noise of an input image to obtain an output image, wherein the H×W pixels of the input image form an H×W input element I ( x , y) of an input array The H×W pixels of the output image form an H×W output elements O( x , y ) of an output array, 0≦x<W, 0≦y<H, and the image noise filtering device comprises An offset array generator that calculates an offset element for each input element I( x , y ), and an offset element corresponding to each input element I( x , y ) represents the input element I( x , y ) and Another input element Difference, 0<p<W, 0<p<H, p is a positive odd number, and when or When the offset element is 0; a difference generating unit is each input element I( x , y ), such that one of the input elements corresponds to the offset element corresponding to the offset element compared to the adjacent input element A difference element is obtained by comparing the offset elements corresponding to the block, wherein the comparison block of the input element includes the input element and a plurality of adjacent input elements; and a calculation unit is an input element I( x , y ), calculating a block element according to the difference elements, wherein the block element represents a sum of all offset elements corresponding to the block based on the comparison of the input elements I( x , y ); An output unit, each input element I( x , y ), based on the corresponding block element and the other input element , the output element O( x , y ) corresponding to the input element I( x , y ) is calculated. The image noise filtering device according to claim 6, wherein, in the case of y=0, when x=0, the difference element calculated by the difference generating unit represents the input element I( x , y The ratio of all offset elements corresponding to the block, when 0 < x < W, the difference element calculated by the difference generating unit represents the ratio of the input element I( x , y ) to all offsets of the block The difference between the sum of the elements compared to the input element I( x -1, y ) for the sum of all the offset elements corresponding to the block; in the case of 0 < y ≦ r, when x = 0, the difference generating unit The calculated difference element represents the ratio of the input element I( x , y ) to the total of all the offset elements corresponding to the block compared to the input element I( x , y -1), and all the offsets corresponding to the block The difference between the sum of the elements, when 0<x<W, the difference element calculated by the difference generating unit represents the ratio of the input elements I( x , y ) to the sum of all the offset elements corresponding to the block compared to the input element differences I (x -1, y) of the ratio of the sum of all blocks corresponding shift elements; in the case of r <y <H when x = 0, the difference to produce a single This element represents the calculated difference between the input element I (x, y) of the ratio of the sum of all elements of the block corresponding to the offset as compared to the input element I (x, y -1) ratio of partial blocks corresponding to all The difference between the sum of the elements, when 0<x<W, the difference element calculated by the difference generating unit represents the ratio of the input elements I( x , y ) to the sum of all the offset elements corresponding to the block compared to the input element The ratio of I( x -1, y ), I( x , y -1) to the sum of all offset elements corresponding to the block. The image noise filtering device according to claim 6, wherein the difference generating unit compares the input elements I( x , y ) with r×r input elements, and The input element I( x , y ) is the upper left corner of the r×r input elements, 0<r<p<W, 0<r<p<H. The image noise filtering device according to claim 8 further includes: a weight processor for each input element I( x , y ), using a corresponding block element N( x ) and the other input element Setting a weight element and a weighting element; and the output unit is a weighted element using each input element I( x , y ) divided by the weight element to obtain a corresponding output element O( x , y ); For the input element of the yth column, the weight processor sets the weight element of the input element I( x , y ) And σ is an adjustment parameter that can control the filtering effect, and r×r is the number of input elements of each comparison block. The image noise filtering device of claim 9, wherein the weighting processor sets a weighting element of the input element I( x , y ) for the input element of the yth column .