Classifications

• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/20—Image preprocessing
  • G06V10/24—Aligning, centring, orientation detection or correction of the image
  • G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
  • G06V30/10—Character recognition
  • G06V30/14—Image acquisition
  • G06V30/1444—Selective acquisition, locating or processing of specific regions, e.g. highlighted text, fiducial marks or predetermined fields
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
  • G06V30/10—Character recognition
  • G06V30/26—Techniques for post-processing, e.g. correcting the recognition result
  • G06V30/262—Techniques for post-processing, e.g. correcting the recognition result using context analysis, e.g. lexical, syntactic or semantic context
  • G06V30/274—Syntactic or semantic context, e.g. balancing
• • G—PHYSICS
  • G06—COMPUTING; CALCULATING OR COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
  • G06V30/10—Character recognition

Definitions

• the present invention relates generally to improved means and methods for automatically recognizing data on documents, and more specifically to improved means and methods for automatically recognizing amount information on financial documents, such as checks, invoices and remittance documents.
• imaging technology as an aid to document processing has been recognized as one way of significantly improving document processing, as disclosed, for example, in U.S. Patent Nos. 4,205,780; 4,264,808; 4,672,186; and 4,888,812.
• imaging involves optically scanning documents to produce electronic images that are processed electronically and stored on high capacity storage media (such as magnetic disc drives and/or optical memory) for later retrieval and display.
• high capacity storage media such as magnetic disc drives and/or optical memory
• document imaging provides the opportunity to reduce document handling and movement, since these electronic images can be used in place of the actual document.
• document images can be retrieved from storage and displayed on workstations where operators can enter amount data and other information based on the observed images, instead of having to view the documents directly.
• Fig. 1 illustrates a typical check of a type widely used in the United States.
• Fig. 2 generally illustrates a document processin system in which the present invention may be incorporated.
• Fig. 3 is a flow chart generally illustrating the various operational steps performed by an automatic courtes amount reader in accordance with the invention.
• Fig. 4 is a flow chart illustrating a preferred manner for accomplishing the "Locate $" Step 102 of Fig. 3.
• Fig. 5 is a typical gray level image provided by the image module 14 of Fig. 2.
• Fig. 6 is a block diagram illustrating apparatus for the parallel generation and storage of seed and mask binary images from the gray level image represented on Fig.5.
• Fig. 7 illustrates a typical "$" seed search area $SA established by Step 102B in Fig. 4.
• Fig. 8 illustrates a typical "$" mask search area $MA established by Step 102B in Fig. 4.
• Fig. 9 is a flow chart illustrating a preferred manner for accomplishing the "Extract Courtesy Amount" Step 104 of Fig. 3.
• Fig. 10 illustrates a typical courtesy amount seed search area C.A.SA established by Step 104B in Fig. 9.
• Fig. 11 illustrates a typical courtesy amount mask search area C.A.MA established by Step 104B in Fig. 9.
• Fig. 12 is a flow chart illustrating a preferred manner for accomplishing the "separate 6 portion and categorize" Step 110 of Fig. 3.
• Fig. 13 illustrates a typical extracted courtesy amount (prior to clean-up in Step 104J in Fig. 9) containing extraneous connected component groups 62 and 63.
• Figs. 14-16 illustrate typical extracted courtesy amounts after clean-up in Step 104J in Fig. 9.
• Figs. 17-18 illustrate how " " characters are extracted from a " ⁇ " field comprised of underlined double figures.
• Figs. 19-21 illustrate how " ⁇ " characters are extracted from a " " field comprised of a fraction.
• the present invention will be illustrated as applied to automatically recognizing the dollar amount (typically referred to as the "courtesy amount") on a check in a document processing system for processing financial documents.
• the present invention is also applicable to other types of documents, as well as to other types of data recognition applications, financial and otherwise.
• Fig. 1 illustrates a check 10 of a type widely employed in the United States.
• the check 10 has a "$" currency symbol 10a, and an associated amount 10b, which is typically referred to in the banking industry as a "courtesy amount.”
• a reader which recognizes this courtesy amount is typically referred to as a courtesy amount reader (CAR) .
• the courtesy amount 10b may be machine printed or handwritten, as shown in Fig. 1.
• the typical check 10 shown in Fig. 1 also includes encoded machine readable data 10c at the bottom-left of the check, which serves to provide identifying information such as the identity of the bank on which the check is drawn, the - 6 -
• Fig. 2 generally illustrates a document processing system in which the present invention may be incorporated.
• the documents to be processed are typically financial documents, including checks of the type illustrated in Fig. 1.
• these financial documents 10 are applied to a document processor 12, which, in a conventional manner, machine reads encoded data from the documents, captures and processes images of the documents, and sorts the documents into pockets (not shown).
• the document processor 12 in Fig. 2 includes an imaging module 14 for capturing images of documents, processing and compressing the captured document images, and then transmitting the compressed document images to storage apparatus 16, such as disk drives.
• Workstations 19 receive document images from the storage apparatus 16 for display and entry of data by workstation operators, such as courtesy amounts from the viewed images.
• a computer processing unit (CPU) 20 provides for overall control of the system, and also for maintaining a data base for document information transmitted thereto by the document processor 12 and workstations 19 (via the storage apparatus 16).
• the document processor 12 of Fig. 2 additionally includes a courtesy amount reader 18 coupled to the imaging module 14 for automatically recognizing courtesy amounts on checks, such as illustrated in Fig. 1.
• a courtesy amount reader 18 coupled to the imaging module 14 for automatically recognizing courtesy amounts on checks, such as illustrated in Fig. 1.
• An important advantage of providing such a courtesy amount reader 18 in the document processing system of Fig. 1 is that those checks whose amounts are successfully read need not have their courtesy amounts read and entered by viewing their images at the workstations 18.
• the courtesy amount reader (CAR) 18 typically comprises a plurality of microprocessors, RAMs, ROMs and other associated circuitry, along with appropriate programming, for operating on document images applied thereto from the image module 14, in order to provide for automatic recognition of the courtesy amounts in accordance with the invention. The manner in which such may be provided for the CAR 18 will become evident from the disclosure herein.
• Fig. 3 is a flow chart generally illustrating the various operational steps performed by the CAR 18 in Fig. 2 in recognizing a courtesy amount on a check. It is to be understood that this flow chart is presented by way of example, and should not be considered as limiting the scope of the invention. For example, certain steps shown herein may be omitted, other steps may be added, and/or the arrangement of the steps may be modified. - 8 -
• the CAR 18 receives a gray level image of a check from the imaging module 14 in Fig. 2.
• the CAR locates the "$" 10a in Fig. 1 (Step 102), and then extracts the associated courtesy amount 10b (Step 104) .
• a determination is then made as to whether the extracted courtesy amount is machine printed or handwritten (Step 106). If machine printed, a relatively simple recognition of the courtesy amount is performed (Step 108) and the result outputed (Step 118). If the extracted courtesy amount is determined to be handwritten (Step 106), a more complex analysis is required.
• the " ⁇ " portion lOb-l (Fig.l) is first separated and categorized (Step 110), and the " ⁇ " characters then extracted based on the categorization (Step 112). The resulting extracted " ⁇ ” characters are then recognized (Step 114).
• Step 116 the dollar characters are recognized.
• the CAR 18 (Fig. 2) then outputs the recognized courtesy amount, or a reject signal (Step 118). In the system of Fig. 2, this CAR output is sent to the CPU 20. If a reject condition is detected during any of the steps in Fig. 3, a reject output is immediately provided and the remaining steps aborted. As shown in Fig. 3, extraction and recognition of the " ⁇ " portion of the courtesy amount are performed prior to the dollar portion, since it is more - 9 -
• the recognized courtesy amount output provided by the CAR can be accompanied by a confidence value based on confidence indications produced during the recognition process. It will also be understood that the recognition Steps 106, 108,
• Step 100 fFi ⁇ . 3 ⁇
• the imaging module 14 in Fig. 2 provides a gray scale image (such as illustrated in Fig. 5) to the CAR 18 of at least the portion of a check containing the B $" character 10a and the associated courtesy amount
• Fig. 5 is by way of example only.
• Step 102 (Fia. 3 ⁇
• the courtesy amount 10b(Fig. 5) is extracted using the previously located "$" character 10a as a location guide. - 10 -
• Step 108 fFi ⁇ . 3
• Step 110 fFi ⁇ . 3 ⁇
• Step 114 the " ⁇ " characters are extracted based on the categorization made in Step 110.
• Step 116 (FL ⁇ . 31
• the CAR 18 outputs (to the CPU 20 in Fig. 2) the recognized courtesy amount, or a reject signal.
• a reject signal is provided by the CAR if a reject condition is detected during any of the previous steps, in which case subsequent steps are aborted.
• a recognized courtesy amount may also be accompanied by a confidence value.
• Step 102 Detailed Description of Step 102
• Step 102 in Fig. 3 A preferred manner for accomplishing Step 102 in Fig. 3, in accordance with the invention, will next be considered with reference to step 102A through 102H in Fig. 4. It will be remembered that the purpose of Step 102 is to locate the "$" character 10a on the check 10 in Fig. 5. Step 102A fFi ⁇ . 4 )
• a thresholding is used to derive a plurality of binary images from the gray level image (Fig. 5) provided by the image module 14 in Fig. 2.
• the derivation of these binary images will be understood by noting that a gray level image may typically be represented electronically as an X-Y matrix of pixels (picture elements), where each pixel has one of a plurality of gray level values. For example, each pixel could be provided - 12 -
• Each derived binary image is produced by employing a different one of these gray level values as a threshold in converting the gray level image to the binary image. For example, if a threshold of eight is used for producing a particular binary image, then that binary image will have black pixels for those pixels whose gray level values are eight or greater, all other pixels of the binary image being white.
• three binary images are derived from the gray level image (Fig. 5) using three different thresholds, high, intermediate and low.
• the high threshold binary image will be referred to as the "$" seed image
• the intermediate binary image will be referred to as the courtesy amount seed image
• the low threshold binary image will be referred to as the mask image.
• the "$" seed image is used for locating the "$" character 10a (Fig. 5)
• the courtesy amount binary image is used for extracting the courtesy amount 10b
• the mask image is used for both purposes.
• the seed and mask images can be generated in parallel by respective converters 34, 36 and 38 as the - -
• gray level image is received from the image module 14 in Fig. 2, the resulting binary images being retrievably stored in respective random access memories (RAMs) 44, 46 and 48.
• Step 102B fFi ⁇ . 4 search areas on the "$" seed and mask images are established for use in locating the "$" character.
• Fig. 7 illustrates an example of a "$" seed image search area $SA for the seed image
• Fig. 8 illustrates an example of a "$" mask search area $MA for the mask image.
• Figs. 7 and 8 also illustrate the effects produced by using different thresholds for deriving the seed and mask images.
• the "$" mask search area $MA in Fig. 8 contains many more extraneous black pixels (noise) than does the "$" seed search area $SA in Fig. 7.
• the desired "$" seed search area $SA in Fig. 7 is known.
• its location could be previously stored in the CPU 20 (Fig. 2), or could be derived from reading the machine-readable line 10c on the check 10 (Fig. 1).
• provision could be made for searching the entire image until the "$" character is located.
• step 102C the "$" seed search area $SA in Fig. 7 is scanned for a "new" black pixel.
• a "new" black pixel is one which has not yet been accounted for in the seed search area $SA.
• Step 102D a new black pixel is not found (Step 102D)
• Step 102E a determination is made (Step 102E) as to whether the last vertical column of the "$" seed search area $SA in Fig. 5 has been scanned. In such case, a reject is produced. It is also to be understood that a reject could also occur if the maximum time alloted for the recognition process has expired. This is done in order to prevent the recognition process for any one check from exceeding a time which would be inconsistent with check processing speed requirements.
• Step 102E If during Step 102E it is determined that vertical scanning has not been completed, operation returns to Step 102C to continue the search for a new black pixel in the scan direction of the "$" seed search area $SA.
• Step 102F If a new black pixel is found during Step 102D, operation proceeds to Step 102F. During Step 102F, the found seed black pixel (Step 102D) in the "$" seed search - 15 -
• Steps 102D and 102F in Fig. 4 take advantage of both of the "$" seed and mask search areas $SA and $MA
• Figs. 7 and 8, respectively to locate and recognize the "$" character. More specifically, when a new black pixel is found in the "$" seed search area $SA in Fig. 7 (Step 102D), the pixel having a corresponding location in the "$" mask search area $MA in Fig. 8 is located. For example, if 34s in Fig. 7 is the new black pixel found in the "$" seed search area $SA (Step 102D), then the correspondingly located black pixel 34m in the "$” mask search area $MA in Fig. 8 is located. This can be implemented using the seed and mask images stored in the respective "$" seed and mask RAMs 44 and 48 in Fig. 6, which may be organized for example, so that corresponding seed and mask pixels have corresponding addresses.
• Step 102F The next operation which takes place in the performance of Step 102F is to propagate the black pixel 34m (Fig. 8) in the "$" mask search area $MA so as to generate a CC comprised of all black pixels connected to 34m. This may be accomplished, for example, using the mask RAM 48 in Fig. 6. Starting with the black pixel 34m (Fig. 8), a determination is made as to whether there are any black - 17 -
• Fig. 8 pixels at addresses corresponding to pixel locations immediately adjacent the black pixel 34m (Fig. 8). A like determination is made for each newly determined black pixel, and then repeated again and again until all connected black pixels forming the CC have been identified. The addresses of these identified black pixels then constitute the CC generated from the black pixel 34m.
• the mask RAM 46 in Fig. 6 may, for example, be used to store the addresses of the identified black pixels forming a CC. Still with reference to Figs. 7 and 8, it will be understood that, if the new pixel found in the "$" seed search area $SA in Fig.
• Step 102D is the black pixel 34s of the "$" character 10a
• the resulting CC produced by propagation of the corresponding black pixel 34m in the "$" mask search area $MA in Fig. 8 (Step 102F) will be CC-1, which is the "$" character 10a. This will be the case since all pixels of the "$" character in the "$" mask search area $MA in Fig. 8 are connected.
• Step 102G operation proceeds to Step 102G where the size, geometry, and location of the generated CC are used to make a relatively fast determination of whether it is an appropriate candidate for the "$" character, or should be rejected, thereby avoiding the relatively more time consuming recognition process.
• Step 102G Only if a CC is determined to be a "$" candidate (Step 102G) will operation proceed to Step 102H where conventional character recognition is performed to determine whether the CC is the "$" character.
• conventional character recognition is performed to determine whether the CC is the "$" character.
• the classifier approach described in the aforementioned U.S. Patent No. 4,449,239 may be employed for recognition. If the CC is not determined to be an appropriate "$" candidate in Step 102G, or if the CC is not recognized to be the "$" in Step 102H, then operation returns to Step 102C to continue scanning for a new black pixel in the "$" seed search area $SA in Fig. 7. However, if the CC is recognized to be the "$" character in Step 102H, then the "$" character has been located. In such a case, no further scanning occurs, and operation proceeds to Step 104 in Fig. 3 to extract the courtesy amount.
• Step 102E if no recognition of the "$" character is made when the end of the scan is reached, then a reject occurs. If it - 19 -
• Step 102F From the foregoing description of Step 102F, it will be ' understood that the "new" black pixel referred to in Step 102C is one that was not previously found as a result of propagation in the "$" mask search area during Step 102F, since there is no need to propagate previously identified black pixels. There are various possible ways of preventing such previously identified seed pixels from being propogated in the "$" mask search area. In the embodiment being described, it has been found advantageous to accomplish this purpose by deleting seed pixels from the "$" seed image (stored in the "$” seed image RAM 44 in Fig. 6) upon identification of the corresponding pixel in the "$” mask search area $MA during mask propagation in Step 102F in Fig 4.
• Step 102F black pixels which were identified during previous propogations in Step 102F are not seen during scanning in Step 102C, thereby reducing the time required to locate the "$" character.
• This savings is in addition to the time saved because the seed image contains relatively - 20 -
• the above described seed/mask propagation approach for generating a CC is additionally advantageous for locating the "$" character 10a on a check 10 (Fig. 1), since the "$" character is normally printed with high quality and high contrast, and is unlikely to produce breaks in the "$" mask search area $MA (Fig. 8).
• submitting each generated CC for recognition, as described above makes it highly likely that the "$" character will be recognized, as compared to other markings or characters (such as the numeral "8" considered previously).
• the seed/mask propagation approach for generating a CC is subject to many variations within the scope of the invention.
• connectionivity used for generating a CC could be changed in various ways to accommodate the recognition of particular types of characters under differing circumstances.
• the definition of "connectivity” could be changed so that connectivity would be restricted to one or more particular directions (such as vertical, horizontal and/or particular diagonals).
• Step 104 A preferred manner for accomplishing Step 104 in
• Step 104 is to extract the courtesy amount 10b shown in Fig. 1.
• Step 104A fFi ⁇ . 9 is to extract the courtesy amount 10b shown in Fig. 1.
• Step 104 (Fig. 9 )
• Fig. 10 illustrates an example of a courtesy amount seed search area C.A.SA
• Fig. 11 illustrates an - 22 -
• C.A.MA in Fig. 11 is of sufficient size to include courtesy amount portions which might project beyond the courtesy amount border 33. Also note in this regard that, even though the "7" of the courtesy amount is not fully contained in the courtesy amount search area C.A.SA in Fig. 10, the "7" will be fully extracted as a result of seed/mask propagation in the larger courtesy amount mask search area C.A.MA in Fig. 11.
• the same mask image stored in RAM 48 in Fig. 6
• the courtesy amount seed image stored in RAM 46 in Fig.
• Fig. 10 illustrates an example of a possible choice of a threshold for the courtesy amount seed search area C.A.SA, wherein the border 33 (Fig. 5) as well as low contrast extraneous pixels (noise) do not appear. In this regard, it is to be understood that all parts of the courtesy amount need not be - 23 -
• C.A.SA IN fIG. 10 It is merely required that sufficient portions of the courtesy amount be included in C.A.SA in Fig. 10 to provide for adequate extraction of the courtesy amount as a result of seed/mask propagation in C.A.MA in Fig. 11.
• Steps 104C, 104D, 104E and 104F (Fig. 9 ) These steps may be generally the same as previously described for respective Steps 102C, 102D, 102E and 102F, in Fig. 4, except that for a normal courtesy amount, there is no reject after the end of the scan (Step 102E), operation instead proceeding to Step 106 (Fig. 3). Steps 104C, 104D, 104E and 104F will thus not be considered in detail. It will be sufficient to note that, each time a "new" black pixel is found during scanning of the courtesy amount seed search area C.A.SA (Fig. 10), propagation in the courtesy amount mask search area C.A.MA (Fig. 11) generates a CC (as previously defined). Step 104G (Fig. 9 )
• Step 104G determines whether the generated CC is likely to be a part of the courtesy amount. For example, a useful basis - 24 -
• a generated CC is a likely part of the courtesy amount is to determine whether it extends to the border 52 (Fig. 11) of the courtesy amount mask search area C.A.MA as, for example, line 55 in Fig. 11. Such a generated CC is most unlikely to be a part of the courtesy amount.
• Step 104G If a generated CC is determined as not likely to be a part of the courtesy amount in Step 104G, then operation proceeds to Step 104H which discards the generated CC; operation then returns to Step 104C to continue scanning for a new black pixel in the courtesy amount seed search area C.A.SA in Fig. 10.
• Step 1041 (Fig. 9 ) If a generated CC is determined to likely be a part of the courtesy amount in Step 104G, then operation proceeds to Step 1041 which stores the generated CC (e.g. in RAM memory 46 in Fig. 6) for later use. Operation is then returned to Step 104C to continue scanning for a new black pixel in the courtesy amount seed search area C.A.SA in Fig. 10.
• Step 1041 stores the generated CC (e.g. in RAM memory 46 in Fig. 6) for later use. Operation is then returned to Step 104C to continue scanning for a new black pixel in the courtesy amount seed search area C.A.SA in Fig. 10.
• Step 1041 Before leaving Step 1041, it will be helpful to note the difference between the way generated CCs are used for locating the "$" character (Step 102, Figs. 3 and 4), and for courtesy amount extraction and recognition. It will - 25 -
• each generated CC is considered as an entity for recognition purposes, since the "$" character is provided with high quality printing and normally has a high contrast and no breaks.
• a CC generated for courtesy amount extraction may be only a fragmentary portion of a character because courtesy amount characters may have several breaks, particularly when handwritten.
• a courtesy amount character may be comprised of a plurality of generated CCs. Accordingly, in extracting and recognizing the courtesy amount, no attempt is made to recognize a generated CC, as is done when locating the "$" character (Step 102H in Fig. 4).
• each CC which is determined as likely to be part of a courtesy amount is stored Step (1041) until the entire courtesy amount area has been scanned, at which time all generated CCs which are likely to be part of the courtesy amount will have been stored.
• These stored CCs then constitute the extracted courtesy amount.
• Fig. 13 is an example of such a stored courtesy amount extracted as described above.
• the seed/mask propagation approach for generating CCs primarily serves as a particularly advantageous way of extracting and storing the courtesy amount for recognition.
• Step 104J fFig. 9
• Step 104J is reached, via Step 104E (which tests for end of scan), after scanning of the courtesy amount seed search area C.A.SA (Fig. 10) has been completed and all generated CCs likely to be a part of the courtesy amount are stored.
• the purpose of Step 104J is to clean up this stored extracted courtesy amount (Fig. 13) by removing extraneous CCs, such as exemplified by 62 and 63.
• One approach used is to delete extraneous CCs, such as 62, if they are spaced a predetermined amount above and below the courtesy amount region. This may be accomplished, for example, by projecting the entire amount field horizontally to define a region having upper and lower boundaries. CCs, such as 62 in Fig. 13, above or below these boundaries are then deleted. If the projection creates a plurality of regions, the appropriate upper and lower boundaries are those corresponding to the region which includes the "$" character.
• C2 The rightmost column of the next area of whit columns with minimum width W left of CI.
• C3 The next column to the left of C2, having a black pixel. If the following condition is met:
• W may have a width corresponding to the width of three columns, the choice of being such that the above condition will not be met by portions of a single character.
• K may typically have a value of 1.5.
• Step 110 is reached if the courtesy amount extracted during Step 104 is determined to be handwritten.
• the purpose of Step 110 is to separate the " ⁇ " portion lOb-1 (Fig. 1) from the dollar portion 10b-2 of the courtesy amount 10b.
• a preferred manner for accomplishing Step 110, in accordance with the invention, will next be considered with reference to Steps 110A through 11OH in Fig. 12.
• the "$" portion and “$" portion of the courtesy amount will hereinafter be referred to as the "$" field and " ⁇ ” field, respectively.
• Step 110A (Fig. 121
• Step 110A the extracted courtesy amount is searched for the presence of a period or decimal point (".”).
• a period or decimal point is, of course, indicative of the separation between "$" and “ ⁇ ” fields of the courtesy " amoun , and its detection can therefore be used as a basis for separating these fields.
• a potential period candidate PC is considered to be an actual period candidate if the following conditions are satisfied:
• the potential period candidate PC has a height which is no greater than 1/2 (UL-LL).
• the potential period candidate PC has a width W which is less than a prescribed amount.
• the average line number of the potential period candidate PC is less than 1/2(UL + LL) .
• Step HOB (Fig. 121
• Step HOB the up to three period candidates determined in Step 110A are investigated using well known statistical classification techniques, as disclosed, for example, in the aforementioned patents. If - 30 -
• Step HOC (Fig. 121
• Step HOC subscripting of the " ⁇ " field of a courtesy amount, such as illustrated in Fig. 15, is investigated as a basis for separation of the "$" and " ⁇ " fields of the courtesy amount.
• the extracted courtesy amount is again investigated from left to right to detect the start of a superscripted character SC (for example, the numeral "5" in Fig. 14).
• Step HOB which describes the search for a potential period candidate
• the upper line UL and lower line LL (Fig. 15) are determined for the courtesy amount portion (such as the numeral "7" in Fig. 15) immediately to the left of the candidate superscripted character SC. Again the lines are numbered from top to bottom.
• a candidate superscripted character SC is considered to be an actual superscripted character if the following conditions are satisfied: - 31 -
• the bottom line number of the candidate superscripted courtesy amount character is no greater than 1/3(2LL+UL) .
• the height of the candidate superscripted character SC is at least 1/3(LL - UL) .
• the candidate superscripted character SC is separated from the courtesy amount portion immediately to the left (such as the numeral "7" in Fig. 15) by at least one white column.
• a white column is a column having no black pixels.
• courtesy amount portion SC (the numeral "0" in Fig. 14) immediately to the right of the candidate superscripted courtesy amount character SC has a lower line number of no greater than the lower line number of SC plus half its height.
• the first candidate which satisfies the above conditions is considered to be the start of a superscripted " ⁇ " field.
• Step HOD (Fig. 121
• Step HOD receives the results of the search for a superscripted " ⁇ " field performed in Step HOC. If a superscripted character was found, a basis for separation of the "$" and " ⁇ " fields will have been determined, and operation proceeds to Step HOG. - -
• Step HOE (Fig. 121 During Step HOE, the presence of a complex " ⁇ " field, such as illustrated in Figs. 16 and 17, is investigated as a basis for separation of the "$" and " ⁇ " fields.
• Fig. 16 shows a first type of complex " ⁇ ” field comprised of two superscripted numerals having an underline.
• Fig. 17 shows a second type of complex " ⁇ ” field in which the " ⁇ " amount is provided as a fraction.
• the values of CI, C2 and LI delineate a particular portion of the extracted courtesy amount for testing as to whether it is a - 33 -
• Step HOE The results of operation in Step HOE are then passed to step 11OF in Fig. 12.
• Step HOF (Fig. 121
• Step HOF receives the results of the search for a complex " ⁇ " field performed in Step HOE. If a complex " ⁇ " field was found, then column C2 serves as a basis for separation of the "$" and " ⁇ " fields, and operation proceeds to Step HOG in Fig. 12.
• Step HOG as a result of having found a basis for separating the " ⁇ " and "$" fields of the courtesy amount, either based on finding the period (Steps HOA and HOB), finding a superscripted " ⁇ " field (Steps HOC and HOD), or finding a complex " ⁇ ” field (Steps HOE and HOF).
• Step HOG provides for separating the " ⁇ " field using the particular basis found for separation (period, superscript or complex " ⁇ " field).
• Step HOG the separated " ⁇ " field is categorized as being one of several types using statistical classifying techniques, such as disclosed in the aforementioned patents. Categories which may be provided by Step HOG for the preferred embodiment being described are double zeroes; double figures, underlined double figures and fraction. Other categories could also be provided.
• Step HOG If an acceptable category is determined in Step HOG, operation proceeds to Step 112 in Fig. 12; otherwise a reject occurs.
• Step 112 Detailed description of Step 112 (Fig. 31
• Step 112 A preferred manner for accomplishing Step 112, in accordance with the invention, will next be considered. It - 35 -
• Step 112 the purpose of Step 112 is to extract the " ⁇ " characters based on the category determined for the " ⁇ " field.
• Step 112 Operation in Step 112 for the various categories provided in the preferred embodiment being described is explained below:
• this category it is immediately known that the value of the " ⁇ " field is zero, and thus operation proceeds to Step 116 in Fig. 3 without further processing.
• this category is used only where the basis for separation is detection of a period or superscripted " ⁇ " field.
• Double Figures For this category the " ⁇ " field figures are directly available so that operation proceeds to Step 114 in Fig. 3 for their recognition without further processing.
• this category is typically used only where the basis for separation is detection of a period or a superscripted " ⁇ " field.
• the slope of the underline is determined as follows. For each column of the " ⁇ " field, the number of white pixels to the first black pixel is counted from the lower edge. If the difference of these numbers for two successive columns is greater in terms of amount than 4, then a position of discontinuity is present. All positions of discontinuity and the greatest column range between two positions of discontinuity in the " ⁇ " field are determined. In this greatest column range, the underline is also expected.
• n straight lines of the slope SL and vertical spacing of 1 is formed.
• the number n of straight lines is dependent upon the slope of the underline and is established as follows: - 37 -
• y_start is established so that (xl yl) occurs under the scanning points of the first straight line.
• y_start is established so that (x2, y2) occur under the scanning points of the first straight line.
• the " ⁇ " field is scanned along these straight lines, with the objective of determining that straight line - 38 -
• the number of scanning points as far as the scanning point with the first black pixel in the " ⁇ " field is counted for all scanning straight lines in the scanning from the right and from the left. Then, the straight lines with the maximum number of counted scanning points in the course of the scanning from the right and in the course of the scanning from the left are determined. From this range of straight lines, that one is selected which is lowest. All portions of the extracted " ⁇ " fields below this lowest straight line are deleted, producing the result shown in Fig. 19. The above procedure also handles the situation where the " ⁇ " characters intercept the underline. After elimination of the underline (Fig. 19) the remaining " ⁇ " field components ("36" in Fig.
• Step 19 are again examined using statistical classification techniques to determine whether it is in a double zero or double figures category. If the category is double zero, operation proceeds to Step 116, since the value of the " ⁇ " field is known to be zero. If the category is double figures, operation proceeds to Step 114 for recognition of the double figures. If neither category is found, a reject occurs. - 39 -
• operation is directed to first removing the denominator, and then removing the fraction line, as illustrated in Figs. 20, 21 and 22 for a " ⁇ " field comprised of a fraction having a numerator "80" and a denominator "100.”
• a preferred implementation for accomplishing this purpose is described below.
• the field is investigated to a column range within which the fraction line is expected. This may be accomplished, for example, by determining the connected component group CC having the greatest width. Once the fraction line has been found, its slope is determined by finding the coordinates xl, yl, x2, y2 and calculating the slope in the same manner as previously described for the underlined complex " ⁇ " field.
• seed and mask images need not be limited to binary (two- level) images.
• a mask image might itself be a gray level (multiple level) image in order to provide - 41 -

Applications Claiming Priority (4)

Publications (1)

Family

ID=25897309

Family Applications (1)

Country Status (5)

Families Citing this family (5)

Family Cites Families (3)

• 1991
  • 1991-09-26 JP JP3517931A patent/JPH06501803A/ja active Pending
  • 1991-09-26 AU AU88532/91A patent/AU8853291A/en not_active Abandoned
  • 1991-09-26 WO PCT/US1991/007121 patent/WO1992006447A2/en not_active Application Discontinuation
  • 1991-09-26 EP EP91918519A patent/EP0551392A1/de not_active Withdrawn
  • 1991-09-26 CA CA002091998A patent/CA2091998A1/en not_active Abandoned

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events