TW569152B - Cutting method of string/character in image - Google Patents

Abstract

There is provided a cutting method of string/character in image, which utilizes the gradient change feature of string in horizontal/vertical direction in digital image and the information of approximate size of the known string to perform alignment of string position. Furthermore, based on the found string position, the gradient and Dilation and Erosion algorithms of Morphology are used to form strings by cutting. After digitizing the string region, the string is made to surround the characters in the string from left to right or from up to down, respectively, and the upper, lower, left and right bounds of each character are defined respectively for dividing the string into independent characters.

Description

L postoperative field] " This post μ is about-a kind of string / character cutting method in the image, special * rainbow type is applied to CCD camera or digital camera or scanner to shoot digital images from 7 Kai Shibao 丨 ^ ^ 子 / / 子 Wu cut side method, using the image processing phase technique 'to cut out the string and character money in the placement image. [Previous technology] 10 At present, the search and search methods for characters of the secret towels are generally based on the gradient method to locate and cut digital images, and then use the upper and lower synthetic contours of the string or vertical / horizontal projection to find the characters. Cut point, 2 strings are cut into separate characters. When the above methods are disturbed by poor digital image quality, skew of the money string, etc., the string / character will often be cut incorrectly. It can be seen that there are still many shortcomings in the above-mentioned customary methods. They are not a good designer, and need to be improved. 15 In view of the various shortcomings derived from the above-mentioned conventional methods, the inventor of the present case, despite thinking and improving and innovating, after years of painstaking and meticulous research, has successfully developed the string / character cutting method in% -cost piece images. [Objective of the invention] The object of the present invention is to provide a method for cutting strings / characters in images, which is applied to a CCD camera or a digital camera or a scanner; The material f looks like a specific string / character search and cutting mechanism to improve the accuracy of Germany. 3 Application for identification of money images [Technical content] 20 569152 A7 B7 V. Description of the invention PA020357.TWP -4/13 10 15 Printed by the Consumers' Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 20… One-word cutting of the image that can achieve the above-mentioned invention purpose Method, using image processing technology for children ’s minds, and using horizontal and vertical gradients of Ding Zao: the characteristics of the transformation and the previously known information about the size of the contract, Perform the nibble of the word position, and then root: the word _ .. j sub ¥ position found, use gradients and Morphology's Dilation and Erosion Yun Zengtai,, Mo, and / 一 才 乃 / ir to perform string segmentation, A After dividing the split string, the upper and lower left and right boundaries of the respective characters are wound from left to right or from top to bottom. The upper and lower boundaries form the character height, and the left and right boundaries form the character width. It is determined that the character width is correct. 1 Each material element is wider than A, indicating that the individual characters may be composed of two or more characters or miscellaneous f, then the upper and lower levels are combined to determine the left and right boundaries of the characters, and according to each wealth Meta: The lower left and right boundaries divide the string into separate characters. In the past, the methods of cutting characters from strings, including projection methods and synthesizing contours, often failed due to the tilt of the characters or impurities between the characters. This caused the strings to be cut incorrectly and reduced the subsequent image recognition. Accuracy, and the method of the present invention is to surround a string of characters in sequence from left to right or top to bottom, and find the upper and lower boundaries of the characters respectively, instead of considering only the vertical or horizontal between the children Compared with the previous technology, it has the accuracy and advancement. > [Brief description of the drawings] 4 With reference to the following detailed description of a preferred embodiment of the present invention and the accompanying drawings, the technical content of the present invention and its purpose and effect can be further understood; the drawings related to this embodiment are: Figure 1 is an embodiment of the character string positioning and character cutting process in the image of the character string / character cutting method in the image of the present invention; please read the precautions on the back before filling in this f guideline-4-this paper size (CNS) A4 X 297 (...) Wang) 569152 A7 V. Description of the invention PAC20357.TWP-5/13 10 Figure 1 shows the string in the image / present; a single image of the string position and mode of the image string method An embodiment of the group process; Figure 2 is an image of the method of cutting the character Feng / Xuanzi ¥ ziwu in this image Φ character cut Wu 1J error, and the embodiment of the process; Figure four is the string / Xuan called Yuping / Ziwu in the image Image of the cutting method _ embodiment of the wrapping method flow of the character cutting module; + Figure 5 is an embodiment of the method of synthesizing the outline of the character cutting module in the image of the string / character cutting method in the image; and Figure 6 T for the image Image towel string character string / character cutting method [Representative symbols of main parts] 1 String position positioning module 11 Input the original grayscale image 12 Divide the original gray image into 9 image areas of the same size, and determine the thresholds I Horizontal / vertical gradient processing Printed by the Property Bureau 1 'Industrial and Consumer Cooperatives 13 Candidates are classified into valid string / rows by using the Diiati〇I ^ Erosion algorithm of Morphohology. 14 Candidate areas based on the previously known string's approximate size and string characteristics. Filter the candidate positions 15 Record the candidate positions that meet the requirements 16 Generate a set of possible word competition Candidate areas 2 character cutting modules (Notes on the back then fill out this page)

Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 569152 βγ PA020357.TWP-3/13 V. Description of the invention (#) 21 The upper and lower right boundary matrix of each character generated by the 21 and the upper and lower strings of the string generated by the upper and lower synthetic contour method The composite contour value matrix 22 cuts the characters according to the upper, lower, left, and right boundaries of each character 23 to determine whether the length and width of the character conforms to the range of previously known characters 24 to change the problematic characters to the upper and lower composite contour values to cut 25 Output all independent characters 41 Input string area image 42 Dynamically binarize the gray area image of the string area 43 A scan line is emitted from the far left of the string (this scan line needs to pass through each character in the string) Detect characters. When the scan line meets the character, set the point as the origin, and then circle around the edge of the character in a clockwise or counterclockwise direction, and record the surrounding position until it returns to the origin. 44 determines the upper, lower, left, and right borders of the character 45 removes the wrapped characters 46 determines whether the characters are all wrapped 47 outputs the upper, lower, left, and right border matrices of all characters 51 input string region image 52 grayscales the character region Image dynamic Binary 53 Calculate the contour value of the string area 54 Calculate the contour value of the string area The paper size applies the Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the precautions on the back before filling in this page)

569152 A7 DA020357.TWP-7/13 V. Description of the invention Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 55 to synthesize contour values up and down 61 sensing unit 62 CCD camera 63 image capture card 64 string positioning unit 65 characters Cutting unit 66 License plate recognition unit 67 Vehicle database [preferred embodiment] Please refer to FIG. 1, which is an embodiment of the image ^ string position determination and character cutting process provided by the present invention, which mainly includes string position positioning Module 1, and character cutting module 2. After the digital image is input, find the most probable position of the string through the 5-position gross position module 1 and cut out the string. Then use the character cutting module 2 to cut each independent word in the string. yuan. Figure 2 shows a detailed embodiment of the process of positioning the string position in the image.输 First, input the original grayscale image 1 and then divide the original grayscale image into 9 image areas of the same size, and determine the thresholds for horizontal / vertical gradient processing12. Calculate the average edge point of the single-region gradient image, and determine the critical value of the effective columns / rows. After all the effective columns / rows are determined and recorded, use Morphobgy's Dilation and Erosion operation methods to classify the effective columns / rows into words. Candidate area 13 of the string area. The 15 candidate regions of 9 image regions are merged, and the generated multiple groups of candidate regions filter the candidate positions 14 based on the approximate size of the previously known strings and the word filial characteristics, and record the candidate positions that meet the requirements

--- (Please read the notes on the back before filling this page) Order ------- ^ 0! ---- .1 H-J ff4-569152 A7

10 15 Printed by the Consumer Cooperatives of the Smart Finance Bureau of the Ministry of Economic Affairs

20 15 ′ finally produces a set of most likely string candidate regions 16. FIG. 3 shows a detailed embodiment of a character cutting module process in an image. Rotate the upper and lower left and right boundary matrices of the respective characters generated by the wrapping method and the upper and lower synthesized contour value matrix 21 of the string generated by the up and down synthetic contour method. First, cut the character 22 according to the right and lower boundaries of the respective subelements, and Then determine whether the length and width of the character conform to the previously known range of character length and width23. If not, it may indicate that the character is connected to another character or impurity due to image noise or binary relationship. Therefore, the characters with _ will be changed by the upper and lower synthetic wheels to cut 24 'and finally output all independent characters μ. -Figure 4 shows a detailed embodiment of the method of wrapping the image towel character cut dragon group. Input the character string area image 41, and perform dynamic binarization of the gray area image of the string area 42 'emitted from the far left of the string—scan line (this scan line needs to pass through each character in the word φ) _character, When the scan line encounters the character, the point is set to '', then wrap around the edge of the character in a counterclockwise or counterclockwise direction, and record the position of the circle until it returns to the original point and a character is wrapped around. After the Yuan ', you can follow the way around!! The upper, lower, left, and right boundaries of the character are 44 ', and the completed character is removed by 45, and then it is determined whether the character is completely wrapped. 4ό If not, 4 is 4, ,, and also by The tracing continues to scan the next character until all the ☆ one hundred Ding Fans in the string area have been wound, and finally the upper and lower left and right boundary matrices 47 of all characters are output.丨 Yueyuji Figure 5 shows a detailed embodiment of the method of the method of combining the contours on the cutting module in the image. The round-robin character ^ order image is dynamically binarized 52;:! Field image 51, the word value is calculated as the contour value 54 under the string area, ^ · ^ area · " Round ^ 53 and the method to find the contour is From left to right, _Home Standard (CNS) A4 specifications (please read the back: Η Η, item, and then fill in this order, 569152 A7 PA020357.TWP-9/13 B7 ί I. Description of the invention (/; ))

J (Please read the precautions on the back before filling this page) In a single line, the detection starts from the top and down. When a string is encountered, the distance from the top to the substring is recorded.丄 When the value of the gallery is 'Lan 7 No', no string is encountered, then the string area south is set to the value of the upper gallery. Therefore, the one-line search method is calculated from the left side of the string area to the right line by line, and the value of the distance 5 obtained by each line is the upper wheel friction value; the same is true for the lower contour. The upper and lower synthetic wheel temple values are the combined upper and lower contour values. The upper and lower synthetic contour values of a single line are the larger of the upper and lower contour values. The largest is the upper and lower synthetic wheel values. If the upper and lower combined contour values are larger than the string area height, the string area height is used as the upper and lower combined contour values. The above single line of action can be calculated from left to right to obtain 10 to the upper and lower combined contour value 55. Printed in Figure 6 by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs is an embodiment of the method for cutting characters in images in the image provided by the present invention, which is an embodiment of an automated license plate recognition process, which mainly includes a sensing unit 61, a CCD camera 62, and image capture. Frame (Grabber) 63, string positioning unit 64, character cutting unit 65, car law recognition unit 66, and vehicle inventory 15. The sensing unit 61 may be a sensor, and the sensor may be a reflection type, a through-beam type, a reflection sheet, or an induction coil (Loop). The system is used to detect whether there is a vehicle in the shooting area. The state change sensed by the unit 61 will trigger an image capture card 63, and the image capture card 63 will capture a single vehicle image captured by the CCD camera 62, and the captured vehicle image will be used for string positioning. Unit 64 locates the license plate, cuts out the license plate, and then cuts the license plate by the character cutting unit 65. Finally, the license plate recognition unit 66 performs the recognition and outputs the complete license plate number; the license plate recognition result can be entered into the vehicle database 67 Whether the comparison is a specific vehicle is used as the management basis for the vehicle access control unit, or the license plate is used to identify the final paper size. The Chinese national standard (CNS) A4 specification (21ϋ X 297 mm) is applicable. 569152 A7

Printed by the Ministry of Economic Affairs, 5 members of the Property Bureau X Consumer Cooperatives V. The invention description (.f) is used as the basis for vehicle toll / flight evasion. [Features and functions] Introduce 4 = The method of cutting strings / characters in the provided images. When comparing the techniques used in face-to-face with each other, it has the following advantages: Elephant 1 position positioning method is characterized by combining shadow name = domain and horizontal / vertical Gradient method and M. # ㈣ 胸 ⑽ E: _ Nose method, to determine the possible position of the word contention on the image: the approximate size and structural parameters of the known strings are analyzed, so the noise on the image ^ minus ^ and '... Reflections and shadows) to improve the accuracy of positioning the guide string. 3: 2nd middle: 'Positioning and cutting out' and then separate the 2nd 2nd 2nd: Come 'The traditional methods include projection method and wheel wide; _ 失效 is invalid due to the inclination of the string or the impurities between the characters , Which makes the ratio of 卞 fan error squares high, and the method developed by this system, the string from the household = or from above, around the character, the upper and left boundaries of the individual characters to find 'ί 不 疋 Early Only the vertical or horizontal relationship between the characters is considered, so Yuyu has higher accuracy and progress than any technology. ^ ^ 3. Words; It is divided into the shadows that introduce impurities in the image noise or dualization. 19 There is a situation where Ding Wu and Zi Wu are connected together. At this time, the two characters cannot be separated in a loop. , But you can use the upper and lower generated by the character :: Anomalies are analyzed in the ancient boundary, and the connection is determined by the upper τ synthesis wheel # method: The local minimum upper and lower synthetic wheel friction .. can assist the ring completion method, Tigu r Accuracy of cutting. Hezi The above detailed description is specific to one of the possible embodiments of the present invention _ 10 5 1〇 20 >）) 0 (Please read the precautions on the back before writing this page)

1 I VI I u n― Stare *-569152

PA020357.TWP .. 1 ^ / 12 Description of the invention (f is clear, but this embodiment is not intended to limit the scope of the patent of the invention, and any equivalent implementation or change that does not depart from the technical spirit of the invention should be included in this case In the scope of patents, in summary, 'This case is not only technically innovative, but also enhances the above-mentioned multiple effects over conventional items. It should have fully met the statutory invention patent requirements of novelty and progress, and filed an application in accordance with the law. Your bureau approves this invention patent tΜ 'to encourage inventions, to the best of your ability. (Please read the precautions on the back before filling out this page) -tj 1 I »·-Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs Applicable to China National Standard (CNS) A4 specification (210 χ 297 public love

Claims (1)

569152 A8 B8 C8 D8 patent application scope [0 printed by the Economic and Intellectual Property Bureau employee consumer cooperatives 20 2 · An image Yin character string / character cutting method, which uses a string to establish a module and a character cutting pattern on the wheel After the eight-digit image, the string positioning module finds the position of the word in the image, cuts the string, and then the character cutting module cuts the string into each independent character. 'The string positioning module The system includes the following steps: Step 1: Input the grayscale image; Step 2: Divide the image into 9 palaces and do the horizontal / vertical ladder processing separately; Step 2: Select the threshold value of each region by region scanning and use Morphoico ^ Dilation and EiOsion methods classify valid rows / effective columns to determine which positions in each zone can contain strings; Step 4: Filter the candidate positions of each zone based on the approximate size and string characteristics of known strings; Step 5: Record the candidate positions that meet the requirements; Step 6: Generate the most probable string standing upright; The character cutting module includes the following steps: Step A: Enter the individual strings Lower left and right boundary matrices and upper and lower synthesized contour values; Step B: Cut characters according to the upper and lower left and right boundaries of the respective characters; Step C: Determine whether the length and width of the characters conform to the previously known range: Find the cutting point to cut the connected characters into independent characters; Step E ··· Output the completely separated independent characters. As described in the scope of patent application, the word in the image " character cutting side 1 (Please read first Note on the back again. 嗔 Copy this page--The order of M ’s line of the book is applicable to the Chinese national standard (cN ^ XTi ^ (21〇χ 297 ^ .: 569152) Patent scope of claim A8 Β8 C3 D8 5 10 3. ： 5: Among them, the lower and right bounding matrix input by the character cutting module is generated by the% winding method flow, and the wrapping method flow can help the following steps: w S Step 1: Input string Area image; Step two: Dynamically binarize the area image of the string; Step two, a scan line is detected from the far left of the string to detect the character, and the point after the encounter is set to the origin, and then clockwise or Wrap around the character boundary in a counterclockwise direction, and record the straight origin of the wrapped position; 疋 Mark the top, bottom, left, and right borders around the character; Remove the characters that have been bypassed; Determine whether the characters are all wrapped; Rotate the top, bottom, left, and right boundary matrices of the characters. For example, if you apply for the image towel character described in Section _, where the character cutting module loses ττ ”= • It is generated by the upper and lower synthetic contour method flow, which includes the upper and lower 80% contour method flow. The following steps: Step 4 Step 5 Step 6 Step 7) > 1 nun —i-^ 1 —Yf ϋ I nr! · (Please read the note on the back? Matters before filling out this page to order ----- line, Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 Steps Steps 2 Steps 4 Steps 5 Steps into the image of the string area; the image of the string area is made dynamically. Elementization of upper and lower synthetic contour values