TW576103B - Accelerative noise filtering method for image data - Google Patents

Description

576103

V. Description of the invention (1) 5-1 Field of the invention The present invention relates to an image data processing technology; in particular, a method for quickly and effectively filtering out image data noise. OFF 5-2 Background of the Invention: Currently, image data encoding is used to perform image data compression, which makes pre-processing and post-processing of image data even more important. In the image data processing technology, there have been some noise-improved methods of image data, such as the median filtering method (medianf17 °), the modified trimming surface average filtering method (MTM filter) (Γ modified trimmed mean filter), FIR-median hybrid filter (FIR-median hybrid filter), and edge preserving smoothing (edge preserving smoothing, etc.) Γ The median filter method (medianfi 11 er) is A non-linear reduction technique that filters impulse noise in an image while preserving edge images. Generally speaking, a two-dimensional median filtering method uses a 3 X containing one pixel to perform noise filtering. 3 or 5 X 5 rectangular mask. In the median noise filtering method, the grayscale value of each pixel is treated like this. Page 5 y / bW3

Page 6 576103 V. Description of the invention (3) '~-^ Select an appropriate mask according to the characteristics of the image area. First, two types of polygonal masks are used, and the degree of change in pixel value of each masked pixel is calculated. Then, the mask of the pixel value change range U is selected to implement the average value transition method. In this way, the purpose of suppressing noise can be achieved 'in the case of image preservation'. However, one disadvantage of this edge / smoothing method is the loss of fine texture (flne texture). Dong Dedong describes the traditional technique of using all the pixels in the selected mask to replace the pixel value of the pixel for which noise data is to be performed. The known method of two-element average is to select the image to perform noise transition = the eight pixels x, y, 2, a, adjacent to the target. , D, level of each-the absolute value of the difference between the pixel value and the pixel value is at a standard deviation value, the adjacent pixel system: the discard value is not more than two different. The absolute value of the difference between adjacent pixels is π i: ^ / '' This adjacent pixel is substituted for the average value, which is too wide. Therefore, there will be b, = aVg (Xyzabcdef), b, = a, two = will encounter the mother as 9, 8, 7 and so on. For example, the average value of a field and a thousand primes is b = avg (xyzabcdef), and the denominator of each adjacent pixel is 9. Therefore, it is not always necessary to use filtering, software, or a combination of hardware and software implementation. The use of more complex computing processes becomes complicated and time-consuming. ,, And calculations will make the page 7 、 Invention description (4) Serve ΐ—a kind of noisy transition of image data ... its filter filtering operation. ,, and can quickly and effectively perform the miscellaneous of the image data ... 3 Purpose and summary of the invention ... The main purpose of the present invention ^ ^ According to the method, i㈣choose η the calculated image data-an audiovisual 4 selection and-center point Pixels (to perform noise filtering, one of the adjacent pixels, or four adjacent pixels in an X-shape relationship, and according to the approximate value of the pixel value of each pixel at the heart point, ... adjacent = adjacent pixel pixel values In the noise filtering operation, the pixel value of the central point pixel can be simplified and / Ϊ replaced. In this way, the noise filtering process of W ^ and shirt image data is processed.

The $ HQ noise filtering method of the present invention: the upper system provides an image data concept that can speed up the calculation, and the eight systems use the concept of binarization (bi nar yi ndex four adjacent pixels) and the center f pixel By forming a cross-shaped or x-shaped relationship, the weighting in the sfL transition operation can be simplified, and the noise filtering operation of the image data can be simplified. The present invention is directed to the purpose of noise filtering and to achieve another purpose. It provides a kind of image data which can accelerate the calculation, which can be used to process the original image data. Page 8 576103 Five 'Description of the Invention (5) The quasi deviation of the image data from the adjacent pixels of the image and the relationship is absolutely two. The value of the two-valued one-prime image noise data is calculated from the large pixel relative value of the four phases of the image, the large pixel relative value, the large value, and the set value. The prime value is as follows: The purpose of the invention is to provide a Add a filtering method. The method of the present invention includes providing $ 2 different selection of one pixel to be filtered as the second two: and selecting a cross-shaped or 'χ $ prime neighbor pixel' from the center pixel in the data. Calculating the center point image The absolute value of the difference between the selected value and the absolute value of the difference is compared with one = small. When the absolute value of the difference is not greater than the standard deviation value, the binary value of one bit should be set to "丨", and when this is the standard deviation value, a corresponding bit of this adjacent pixel is determined as " 0 ", so as to obtain a 7C mapping table containing a four-bit of the four adjacent pixels. The bitmap table deduces that (I) 'S has a different pixel value to replace the center point image, such as Xο Σ KiXi Σ Ki (i) /, where X〇 represents the pixel value of the center point pixel and W represents the new Xi (1 = 1 ~ 4) represents the pixel value of each adjacent pixel, χ belongs to the scan line before the scan line to which the central point pixel X〇 belongs, and ^ and χ3 belong to page 9 576103 V. Description of the Invention (6) The center point pixel X. The scan line after the scan line to which it belongs belongs to a good scan line X4 belongs to the center point pixel X and any of these adjacent pixels—Ge Xi 1 1 _ ~ 4) as the center When the binarization value of a dot pixel is set to " 〇 ", Dragon 4 uses a weight coefficient, and when the binarization value of a neighboring pixel is set to & q uot; 1, where the two σ ^ coefficients K are 0, and have the bits of the present invention to filter, remove shadows * data miscellaneous 目 ^ pixels have a weighting coefficient of 才 目 # pixel pixel value is similar to the pixel value The noise of this center point pixel has passed itH Risi's weight = 疋: An adjacent pixel performs the weighting of the pixel value of the center point adjacent pixel flat f "· 'and replaces the center with the center point pixel and the four-phase pixel value. Point = 1 Jht'd —age), find :: New and accelerated noise of the image data refined method of the present invention 'simplifies the purpose of the present invention and customers are very TOI, 1 B day „A ^ Ju Xi advantage by The detailed description of the following specific embodiments will be made clear with reference to the accompanying drawings. 5-4 Detailed description of specific embodiments. The present invention provides a method for filtering image data that can accelerate calculation. The method k selects a target pixel pixe㈠ (to perform noise over L) into ten subtypes or X. Four adjacent pixels in a font relationship, and based on the similarity between the pixel values of these neighboring pixels and the center point pixel, further select appropriate neighboring pixels from the four neighboring pixels to perform noise filtering operations. computation) to obtain a new pixel value in place of 576103 Five, Invention Description (7) Heart image | I & uses μ as pixel value. This new pixel value is the center pixel pixel value and the generation ... °, the weighted average of the pixel values of adjacent pixels (weighted = 2 § :: f) The method of the present invention can simplify the noise filtering operation to only

Sh 1 ft) and add (add) mode, that is, only the shift register 彳 f + ·, 丨 is used on the hardware, and it is only used to make the 1) register and the adder. (Adder). It ’s so simple. [可 明 “: 口: Noise data calculation is better than the traditional technology by the following preferred embodiments. In conjunction with the drawings, the ΔΗ and Β diagrams are a preferred jl if according to the present invention. f ^ > Flow chart of image data noise filtering method, U =, for image data to choose from this image data-mid; point; i2 ('a targetplXel), which is one of the noise filtering to be performed (at〇 be fUtered). Next, in step 103, see the second and P. The adjacent pixels are adjacent pixels pi, P2 3 and P4. The pixels P! belong to the center point like Xin p. 1 2 ^ ^ ^ p 2 ^ p 3 ^ ^ ^ ^ The knowledge of which belongs, 'where the adjacent pixel p2 is J. Is the scan line pixel h after the scan line to which the central point pixel PQ belongs. Next, in step 104, calculate the central point pixel p G and each pixel

Page 11 576103 V. Description of the invention (8) is the absolute value of the difference between the pixel values of the pixels 1V2? 3 ~. At step =, compare mother:? selected? The magnitude of the absolute value of the difference between adjacent pixels and the-standard and value. When the absolute value of the difference is not greater than the standard deviation, the binary value corresponding to one bit of this adjacent pixel is set to 1 ”(step 1 0 6). When the absolute value of the difference is greater than the standard deviation 'The binarized value of the corresponding one bit of this adjacent pixel is set to " 〇 " (step 07). Then, in step 108, based on the results generated in steps 105 to 107, a four-bit value is established to include the four A 4-bit mapping table of the binarized values of the neighboring pixels P !, P2, ρ3, and P4. Refer to Table 1 shown in the third figure, this four-bit mapping table It satisfies one of the sixteen examples shown in the table. That is, the four-bit binarization value combination of the four neighboring pixels prp 2, p 3, and p 4 can be sixteen as shown in table 1. Either of these cases, namely Example 1 to Example 15. Based on this four-bit mapping table, a calculation & work is performed, and in step 10, formula (I): Σ KiXi

Xo1—- Σ Ki i = 0 where (0 X〇 represents the pixel value of the center point pixel and XG, and represents the one that wants to replace X◦

Page 12 576103 V. Description of the invention (9) '~: new pixel value, Xi (i = Bu 4) represents the pixel value of each selected adjacent pixel, X !, χ2, χ3, and χ4 are adjacent Pixels P1, P2, D2, and P4 ^ pixel values, K i (1-0 ~ 4) is the weighting coefficient of the center point pixel p and any of the four adjacent pixels p, P2, P3, and P4 (weighting When the bit binarization value of one of the neighboring pixels is set to ^ ", its weighting coefficient Ki is 0, and the bit binarization value is set, and these neighboring pixels of 1 have equal weighting. Coefficients. The first table shown in the fourth figure, „, the members do not map to the center point of each example of Table 1, as shown in Example 1 of Table 2, when each; two weight coefficient combination relationship. The weighting factor K of the point pixel, ... is the value ".", P, P, and P * / number system 4 丨. When the four adjacent pixels p 丨, p2, p3, and P4-two of adjacent pixels Gan ^ The binarization value of the neighboring pixels is "quote.", When the middle 1 'the remaining three & " 1 "Xishi * Rime κ 7 multi-primary γ ^ point pixels and the binarization value weight of this neighboring pixel The coefficients are all values two, four, and eight. When the four adjacent images are recorded as example one, the pixel binarization value is " r, and the rest, one, one, two, and only two and three adjacent ones. / The weighting coefficients of the adjacent pixels are 1/4 and the weighting coefficients of the intermediate pixel P0 are five, six, nine, ten, twelve, and thirteen. 〃 录 一 之Example 2: The value of ννΛ adjacent residual pixels is wide, P2, P3, and the middle-adjacent pixels-the value of the two adjacent pixels is 0 and two of the eight adjacent pixels. The value of the selected value is the value of "1". No. 1 in the second j day * when the value of 1 is selected, the weighting coefficients of these two adjacent pixels are all, two 576103 five plus, invention description (10), and sum coefficient is 1/2, see Table 2 For example, Table 2 'is better than having binarized values ^ 7 ^ 1 and 14. Refer to Table 1. The center point pixel belongs to the same scan = the selection formula U in three adjacent pixels. Two adjacent pixels are substituted into the calculation. 'The two adjacent pixels p, one adjacent pixel p 2, P3, and p 4 are selected. The two adjacent pixels p 2 and p 2 3 are substituted into the calculation formula (I), and The weighting factor is 1/2. "The H coefficients of the image data are all" 4 "and the center point pixels are often captured along the stepping scanner that drives the scanning head, which leads to the phenomenon of biurring. Pixels in this direction are easier to generate, and P2, Pa, and P4 are three adjacent ^. When the four adjacent pixels are p, 1 ", it is better to choose two adjacent to 1, h and P4 have binary values, and the two neighboring pixels p: 3 and p 4 are substituted into the calculation formula (I). The weighting coefficient of the pixel P 0 is 3. The weighting coefficients of both 4 are 1/4 and this. Four adjacent pixels p 1, p 2, p. Furthermore, as shown in Reference Example 14, when P4 has a binarized value "^", 2, 3 and middle three adjacent pixels P !, P2 are substituted into the calculation formula (I), and ^ is better The two adjacent pixels Pi and P2 can be selected to be 1/4 and the center Z pixel p. The weight of the two adjacent pixels P! And P2 is that the adjacent pixels Pr p2, p3, and ^ G t have a weighting factor of 1/2. For these four 'T's, the weighted coefficient of each-adjacent pixel is the weighting coefficient of the binarization value of the adjacent pixel is i / 2, for example, ^ 糸 is 1/8 and the center point pixel P 0 〆n Ding Wu. Next, in step 1 1 0, the calculated pixel value X is calculated by calculating the new // value of this four-bit mapping table with v. . Next, replace the center point pixel at "χ." ~ Step 1 1 1 to determine whether the next-pixel is

Page 14 576103 V. Description of the invention (11) Noise filtering computation is to be performed. To continue the noise filtering operation of the image data, proceed to step 1 1 2, shift to the next pixel, and repeat steps 10 2 to 1 1 1. To end the noise filtering operation of the image data, proceed to step 1 1 3. According to the above, the present invention only needs to use a shift register and an adder on the hardware to perform image data noise filtering operations. Therefore, the method of the present invention can reduce the complexity of the hardware architecture, and whether the method of the present invention is implemented on hardware, software, or a combination of hardware and software, can significantly reduce the computing time. Referring to FIG. 2B, in another preferred embodiment of the present invention, four adjacent pixels P r P 2, P 3, and P 4 in an X-shaped relationship with the center point pixel P 0 are selected. In addition, the image data noise calculation steps in this preferred embodiment are the same as steps 101 to 113. The above are only specific embodiments of the present invention, and are not intended to limit the scope of patent application for the present invention; all other equivalent changes or modifications made without departing from the spirit disclosed by the present invention should be included in the following Within the scope of patent application.

The 576103 diagram on page 15 briefly illustrates that the first diagram A and the first diagram B are flowcharts of steps in a method for filtering image data noise according to a preferred embodiment of the present invention; the second diagram A illustrates the comparison of the first diagram A mask selected in the preferred embodiment; the second diagram B illustrates another mask selected in another preferred embodiment of the present invention; the third diagram is the noise of the image data of the present invention A four-bit mapping table used in the filtering method; and the fourth diagram is a weighting coefficient combination relationship table of a central pixel and its four adjacent pixels in the image data noise filtering method of the present invention. Main symbols: 1 0 1 ~ 1 1 3 Process steps of image data noise filtering method Page 16

Claims (1)

576103 6. Application patent scope 1. An image data noise filtering method capable of accelerating calculation, which includes providing image data; selecting one pixel of the noise to be filtered as a center point pixel from the image data; and from the image data Select four adjacent pixels that have a cross-shaped relationship with the center point pixel; calculate the absolute value of the pixel value difference between the center point pixel and each of the adjacent pixels; compare the absolute value of the difference with a standard deviation, When the absolute value of the difference is not greater than the standard deviation, the binarization value of a corresponding bit of the adjacent pixel is set to π 1π. When the absolute value of the difference is greater than the standard deviation, the adjacent pixel The corresponding binarization value of one bit is set to π οπ to obtain a four-bit mapping table containing the binarization values of the four adjacent pixels; and according to the four-bit mapping table A formula (I) is derived to calculate a new pixel value to replace the pixel value of the pixel at the center point: Σ KiXi V, i = 〇AO = / 、, the scope of patent application is in the middle and middle, which represents the value of the center w, L α = ι ~ 4), the pixel value of the pixel and Xfl, which represents the new image The center point pixel X mother-the pixel value of the adjacent pixel, X 丨 belongs to | It belongs to the scan line before the scan line at the center point de i, L and the scan line to which the pixel xG belongs , L is a scanning line after going to the center point / tracing line, K i (i = 〇 ~ 4), when the neighboring pixel is a weighting coefficient of any adjacent pixel such as 等 °, the coefficient K i When the value is 0, and the value is set to ", the adjacent pixels of the weighting and so on are present; the value of the two f 4 bits is set to, and the one of | 1 " has a weighting coefficient of phase 4. 2 · If you apply for the No. 1 noise data method in the scope of the 5th Patent, where the above =: calculated image data ", the weighting coefficient of the center pixel: is the binarized value of the neighboring pixels ". 3. The 1st jg Zhang Noise transition method in the scope of the 5th patent of Rushen, in which the binarization value of the above-mentioned four-phase image data is "r" and the center point of "$ adjacent pixels of the remaining three pixels" The pixel and the binarized value of the binarized value are " 〇 " when the coefficients are both 1/2. The weight of the adjacent pixel is 1. The noise filtering method as described in item 1 of the patent application range, where The above-mentioned four phases: = The binarized value of the image data calculated by the speed calculation is " 丨 " and the remaining two elements are the binarized values of the neighboring pixels and the neighboring pixels are " 〇, 18th Page 576103 〕 / 61 〇3, patent application scope ___ a noise filtering method, 1 φ '^ value is "〇" and one of the pixels adjacent to the special adjacent pixel 1, choose to have two letters ;: The remaining two f # the binarization of neighboring pixels is the direction of the movement of the knife — # 松 汗, who is one step of the scanner, who is-one of the Xitang, the neighboring pixels are substituted into the nephew + γ τ ,, A—The adjacent image turbulence 4 ρ 尨 th th th 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 彳 for different formula (ί) operation, I's weighting coefficient is 1/2. ', Σ 隹, and white are 1/4 and the center point pixels | including ... can speed up the different image data noise filtering methods, which includes image data; select one of the noises to be filtered from the image data The child goes to the point of the pixel; the pixel as a value of a central relationship must be selected from the image data and the four adjacent pixels in the image; ten. The pixel at the center point and each adjacent pair of values; and the absolute value of the difference between the pixel values is set to " 丨 ", when the absolute value of the difference is greater than two values of the standard bit, The binarization value of the corresponding one bit of the adjacent pixel is set to "quota", and the value can be obtained from a table containing the binarization values of the four neighboring pixels; and the absolute value of the difference is compared with A standard deviation value. When 誃 is not greater than the standard deviation value, the adjacent image Dong * 丄 is used to obtain a four-bit map. A ancient 4 · 古, τ J 〇 derived from the four-bit map. Ding Yi (I), a new pixel value to replace the pixel value of the pixel at the center point. ° T 异 576103 VI. Patent Application Range Σ KiXi Xo '—- Σ Ki i = 0 Where X 〇 represents the pixel value of the center point pixel and X 〇' represents the new pixel value, X i (i =: 1 ~ 4) Representing the pixel value of each adjacent pixel, X i belongs to the scan line before the scan line to which the central point pixel X 〇 belongs, and X 2 and X grandson belong to the central point pixel X. The scanning line to which X 4 belongs belongs to the scanning line following the scanning line to which the central point pixel X 0 belongs, and K i (i = 0 ~ 4) refers to any one of the central point pixel and the adjacent pixels. Weighting coefficient, when the binarization value of the bit of the adjacent pixel is set to " 0π, the weighting coefficient K i thereof is 0, and the binarization value of the bit having the bits is set to π Γ The adjacent pixels have equal weighting factors. 11. The method for accelerating calculation of image data noise filtering according to item 10 of the scope of patent application, wherein when the binarization value of each adjacent pixel is π 0π, the weighting coefficient of the center pixel is Is 1. 1 2. The image data noise filtering method with accelerated calculation as described in item 10 of the scope of patent application, wherein one of the four adjacent pixels described above is the adjacent pixel On page 21, the binary value of the adjacent 576103 6. The binary value of the patent application range is `` 1 '' and the remaining three adjacent images are 'the pixel of the center point and the binary value is 1' , And the weight coefficients are all 1/2. 1 3 · The accelerating noise filtering method as described in item 丨 0 of the patent application scope, wherein the value of one of the two primes of the four adjacent pixels is π 1 ”and the other two adjacent pixels, π 0 ", the binarized value is " 丨", the two adjacent pixels are 1/4, and the weighting coefficient of the center point pixel is 1 / 2. 14. The method for accelerating the calculation of miscellaneous materials as described in item 10 of the scope of the patent application, wherein the binarization value of one of the four adjacent pixels is "0", and the remaining three such When the neighboring pixels are 1, two of the binarized values " 1 π are selected to be entered into the calculation formula (I) operation, where the sum of the two neighboring pixels is 1/4 and the center point The weighting coefficient of a pixel is i 2. 1 5 · The accelerated noise filtering method as described in item 4 of the scope of the patent application, wherein the two adjacent pixels selected to be substituted in the above belong to this method. The value of the point pixel ^ belongs to ^. The pixel value plus "image data" and other adjacent image values are equal to 7 weight coefficients. "Image data $ and other adjacent images' value are 3 adjacent pixels." Scanning lines of image data (I) calculation lines with all weight coefficients. Image data of the adjacent image 576103 VI. The scope of patent application is 1/8 and the weighting coefficient of the center point pixel is 1/22. 7 · Accelerated calculation of image data noise filtering method as described in item 10 of the scope of patent application, in which the above-mentioned color image The data is a stroke = take. ~ Acquired by the scanner 18 · The accelerated noise filtering method as described in item 17 of the scope of the patent application, wherein the above two adjacent pixels and the binarization value of the image element are "0" And the remaining three such phases—when the adjacent images are Π 1 Π, the two phases with the binarized value " 1 "are selected to enter the motor moving direction. These phases are not along the scanner. One step, in which the addition / prime of two adjacent pixels is substituted into the calculation formula (I). The weighting coefficient of the operator is 1/2. The coefficients are all 1/4 and the center point image.